👁 Image Archives

Archive 1 Archive 2 Archive 3 Archive 4 Archive 5 Archive 6 Archive 7 Archive 8

This page has archives. Topics inactive for 30 days are automatically archived 1 or more at a time by Lowercase sigmabot III if there are more than 4.

• BrightOrion (talk · contribs)

This user seems to have started using AI at some point, but has contributed since before the widespread access to LLMs. I've draftified some of their newer creations that I could identify as being generated, but they've created a lot more articles. Would appreciate it if someone else can take a look and help slog through this. --Gurkubondinn (talk) 09:50, 30 March 2026 (UTC)[]

The metaphorical smoking gun is in Diff/1346161779. --Gurkubondinn (talk) 09:52, 30 March 2026 (UTC)[]

Despite your warnings, they're still using LLMs, comment, but are yet to make a proper mainspace edit since. Worth keeping an eye on Kowal2701 (talk, contribs) 22:28, 1 April 2026 (UTC)[]

Looks like they have started now though. A quick glance at their recent contribution history lead me to Special:History/William and Mary (1817 ship), with some very rapid-fire edits. --gurkubondinn 16:57, 1 June 2026 (UTC)[]

Yup. Their latest adds "Daniel Scholes" to the winner list in Poohsticks, cited to an article that does not mention him, though it does mention the game. Dated May 26. We need not to just sit on this one. M kuhner (talk) 02:40, 16 June 2026 (UTC)[]

This user has WP:AUTOPATROLLED, we should definitely not sit on this. ‑‑gurkubondinn 09:30, 16 June 2026 (UTC)[]

I've notified them on User talk:BrightOrion § AINB notice, let's see what they have to say. In the meantime, I'll draftify their two latest creations (one of which has a CIRCULAR citation to commons). ‑‑gurkubondinn 09:34, 16 June 2026 (UTC)[]

Mosherdude91 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I think this user may be using LLM to "improve" and generate content. The speed at which they churned out articles in March and April 2026 is just too hard to dismiss (see their edit history); the average editor can't possibly produce this much high-quality content at such a rapid pace. For instance, in March 2026, they made a massive improvement on the Gigli article on March 21, just three days after having made another "major write" in another article, March 18, 2026. The latter article got reverted by another editor for having "too many broken or malformed references" which, to me, looks to be AI-generated due to the naming scheme of the repeated refs (e.g., "BTG-CheapFlights-2012"). In April, user made a "Major overhaul for FA-readiness" of the Heartstopper (TV series) article on April 2, which was followed by an expansion of List of Heartstopper episodes for a potential WP:FLC on April 10. User's other content may have to be checked for signs of AI writing. Nineteen Ninety-Four guy (talk) 08:51, 26 April 2026 (UTC)[]

Hi Nineteen Ninety-Four guy, thanks for this notice. Until recently, I had been reading https://en.wikipedia.org/w/index.php?title=Wikipedia:Large_language_models&oldid=1344513727 and https://en.wikipedia.org/w/index.php?title=Wikipedia:Writing_articles_with_large_language_models&oldid=1345232327 as a guideline for editing wiki articles. Whenever I used AI to edit, I always made sure to verify the integrity of what was used, including reading the source to make sure that the text was supported by correct sources. However, I accept that my process was not good enough in some cases. In the case of the Fascinating Aida article, my initial version had malformed sources as a result of me not previewing it properly. When I became aware of the latest policy, which appears much stricter, I stopped using it in this way. However, even under the old guidelines, my use of AI to improve articles may have breached the core principles set out, which is "Avoid using large language models (LLMs) to write original content, generate references, add citations or create replies on discussion pages".

Moving forward, I will no longer use LLMs to write any article content at all. I apologise for any rules I've broken. I will no longer do so, and am happy to give any further explanation necessary. My aim has always been to genuinely improve Wikipedia for the content I care about, but I do not wish to breach policies and rules to do so.

That said, are the following uses acceptable under the guidelines?

- Using AI to find sources that may be useful for an article, ensuring that I write the prose myself

- Using AI to correct errors such as spelling/grammar, without using it for full rewrites

- Using AI to format wikitables, ensuring all the content of the tables is human-written

I appreciate any and all guidance. Mosherdude91 (talk) 11:52, 26 April 2026 (UTC)[]

1. Yes though you must be careful to read the source yourself and not ask AI to summarise it for you. Also go through the source urls and remove the utm?source=chatgpt.com to avoid tripping edit filters.

2. Kind of. If it really only fixes spelling and grammar then probably but in practice there is great difficulty in getting an LLM to stick solely to that. The relevant exemption is Wp:Basic copyediting.

3. Strictly speaking no but if that is really all the LLM does it is unproblematic and nobody will be able to tell. Again, in practice there is great difficulty in restraining the LLM.

Fermiboson (talk) 12:05, 26 April 2026 (UTC)[]

Per WP:NEWLLM, yes they are. But you are being a little evasive and disingenuous here, quite frankly. Am I to understand that you used AI merely to locate sources and fix grammar in the Gigli article? Looking at that article's Production and Marketing section, I see an abundant use of in-text attribution where it's not really needed, something attributable to AI writing:

• Tim Robey wrote that Roth initially aimed to keep Revolution's films below $50 million, but Sony's commercial rebound in 2002 encouraged a more expensive 2003 slate; after several costly underperformers, including Gigli, Revolution lowered budgets, curtailed back-end deals and reduced staff.
• In 2003, the L.A. Times reported that Brest, like Ron Shelton, had been allowed to write, produce and direct his picture with final cut, an arrangement Revolution later reconsidered after those underperformers
• Variety reported that Lopez received $12 million for the film; Robey added that this was $500,000 less than Affleck's reported salary and that her deal also included back-end participation in the high single digits.
• The production notes also state that Gigli's apartment exterior was based on a Hollywood building, while the interior was constructed on a soundstage.
• The L.A. Times also reported that Sony briefly considered renaming the film Tough Love because executives feared audiences would mispronounce Gigli
• According to people involved in the film's release, Sony and Revolution struggled to market Gigli against "a tide of bad press", while the off-screen relationship between Affleck and Lopez became the picture's "greatest marketing liability".
• Contemporary reporting attributed the film's weak commercial performance

The fact is that in-text attribution should only be used to describe subjective or exceptional claims (e.g., Roger Ebert said that the film was "unforgettable".) Furthermore, it's rather odd that you would use naming schemes such as "Guardian-CheapFlights-2012" and "WOS-CheapFlights-2011" to call out repeated refs in the Fascinating Aida article; again may be a sign that you've been using AI beyond source locating and grammar fixing. Nineteen Ninety-Four guy (talk) 12:26, 26 April 2026 (UTC)[]

I'm not trying to be disingenuous, and I apologise if my previous response looked that way. To clarify, you are correct: I did not use AI merely to locate sources and fix grammar. In some of the article work I've done, including Gigli, I used AI in some of the drafting and rewriting of the prose, which went beyond the guidelines of AI use on Wikipedia. That was my fault entirely for not adhering closely to the rules, and I'm not trying to evade responsibility. I'm happy to help identify and repair any edits I've done using AI this way. And if editors think that an article I've edited should be reverted rather than repaired, I have no objection. Once again, my intention was only to improve the articles, but I understand that the process matters as well as finished work itself. Mosherdude91 (talk) 14:21, 26 April 2026 (UTC)[]

Well, that's for the regulars on here to decide whether the best course of action is a cleanup or an outright ban from editing due to persistent AI misuse (provided you've been warned). Nineteen Ninety-Four guy (talk) 14:38, 26 April 2026 (UTC)[]

Blocks are preventative, not punitive. Fermiboson (talk) 15:07, 26 April 2026 (UTC)[]

WP:Mistakes are allowed - now that Mosherdude91 is aware of the expectations and has expressed a commitment to follow them and to help clean up any mess they've already made, I'm happy to let them do so. Of course, continued LLM use would be viewed dimly at this point. -- LWG ^{talk (VOPOV)} 15:58, 26 April 2026 (UTC)[]

OK, fair enough. Shall Mosherdude curate a list of articles they have written with AI to provide a starting point for the cleanup drive? Nineteen Ninety-Four guy (talk) 16:42, 26 April 2026 (UTC)[]

I've created User:Mosherdude91/Review log to help me, and any other editors who wish to, track and review my recent edits where I have used LLM assistance in ways that may not conform to WP:LLM. My intention is to go through these articles and repair or revert any affected changes as needed. Mosherdude91 (talk) 01:20, 28 April 2026 (UTC)[]

@Dr vulpes: One of Mosherdude's LLM-assisted contents, The Survivalist (2015 film), has just been promoted to GA by reviewer Vestrian24Bio despite Mosherdude's LLM declaration near the latter part of the review. Was the reviewer well within their rights to do so despite the article having been wholly or partially generated by AI? Because I feel like a cleanup should have taken place first during the review. Nineteen Ninety-Four guy (talk) 12:34, 28 April 2026 (UTC)[]

As far as I've checked I didn't notice any LLM-related problems in the content.. Vestrian24Bio 10:55, 29 April 2026 (UTC)[]

Evidently you didn't look closely enough. I am in the process of finding and correcting several AI hallucinations in that article. Helpful Raccoon (talk) 21:37, 29 April 2026 (UTC)[]

@Dr vulpes: I noticed that articles listed on User:Mosherdude91/Review log has been tagged with Template:AI-generated. I understand why that is, and I'm not asking for it to be taken down while the article still needs cleanup. I'm also not requesting that this section of the noticeboard be taken down or hidden either, as I understand the importance of this cleanup. I also understand that, as per Template:AI-generated, it's permitted to link to this "user" section of the noticeboard. However, I did a spot-check on other AI tagged articles such as Sex and the City, George O'Malley, Sridhar Makam Krishnamurthy and Fire Island (film), none of which include any usernames. As you marked my case as "committed", I feel like my username appearing on each tag gives an unnecessarily harsh impression of the noticeboard status, and makes personal what should be an article-maintenance issue.Mosherdude91 (talk) 03:21, 29 May 2026 (UTC)[]

• Antonio de Casmedi (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I just got done with draftifying 52 articles from this user (should maybe request the pagemover userright to save admins the time from cleaning up redirects), created rapidly over the last five days.

Most of them were created with {{multiple issues}} tags and tagged as 'possible cut and paste move or recreation'. A few were even created with [citation needed] tags inline, which I always find funny. These may be unattributed machine translations, since most of the articles also existed on idwiki (linked through Wikidata). Would appreciate a second set of eyes on this.

I did not draftify Agra Mas and only tagged it, because Draft:Agra Mas already exists, with a near-identical text. The draft is shorter, and created by a different user, so I'm not sure what's going on there. --Gurkubondinn 12:30, 2 May 2026 (UTC)[]

Hi, thank you for your support. Some of my articles seem to have been moved to draft. I will work hard to improve these articles until they are ready for publication. However, if I use a translator, is that permissible? Many of the articles I use as references are not in English.

Some of the articles that appear to be written too quickly are because I wrote them first and then saved them in my personal archive. I did not use AI to create these articles. I did use AI for some reference searches.

Thanks. Antonio de Casmedi (talk) 12:48, 2 May 2026 (UTC)[]

Hi Antonio de Casmedi, do you mean can you use a translator to read sources in various languages? You can, with the general awareness of the shortcomings of any machine translation. If you mean directly translate articles from other language Wikipedias, that is not permitted per WP:MACHINETRANSLATION.Agra Mas appears to be an unattributed translation of id:Agra Mas. Draft:Agra Mas appears to be a partial translation of some kind, it seems to have taken the structure and some prose but not all prose and with some new prose. CMD (talk) 13:36, 2 May 2026 (UTC)[]

Hi Gurkubondinn, okay I understand.

Yes, that's what I meant. Because some of the sources for the articles I wrote were taken from non-English sources, and I rewrote them in English and used those sources as references.

Regarding Draft:Agra Mas, I didn't realize that the draft already existed, and I didn't check it. I have to relearn how to write perfectly.

But if I improve some of those drafts, will they be published?

Thanks Antonio de Casmedi (talk) 13:59, 2 May 2026 (UTC)[]

The timing will depend on the particular draft, but why did you create them with maintenance tags? CMD (talk) 14:02, 2 May 2026 (UTC)[]

Oh, sorry, maybe I was mistaken. By using the maintenance tag, I thought it would help other users notice a new article and make it more perfect.

So, what tag should I use?

Coincidentally, I'm also new to creating an article; previously, I'd only edited or added new content to a few articles. Antonio de Casmedi (talk) 14:26, 2 May 2026 (UTC)[]

Understandable. For translations, there should be an attribution in an edit summary like the example at WP:TFOLWP. That is important for legal reasons, but it also helps other users understand where all the text came from. Machine translation risks a few errors, for example your articles seem to have translated the reference titles from Indonesian to English. This is getting a bit off topic for this noticeboard, but if you have questions WP:TEAHOUSE is a good place to ask. Working a bit slower is also advisable, get some quality articles out. CMD (talk) 14:38, 2 May 2026 (UTC)[]

Okay, thank you.

I have one more question.

If I write an article in the future, should I put it in drafts? This is so that other users can review the article before publishing it. Antonio de Casmedi (talk) 14:51, 2 May 2026 (UTC)[]

If you want a review, you should create a draft using the Wikipedia:Article wizard. CMD (talk) 14:58, 2 May 2026 (UTC)[]

Okay, understood. Thank you for your support. Antonio de Casmedi (talk) 15:01, 2 May 2026 (UTC)[]

@Antonio de Casmedi, you're of course welcome to continue editing here, but you may find it easier to contribute to Banjarese Wikipedia, Javanese Wikipedia, or Indonesian Wikipedia? Kowal2701 (talk, contribs) 18:32, 2 May 2026 (UTC)[]

@Dieter Lloyd Wexler: you might want to be aware of this, I saw your notice on the user's talk page just now. --gurkubondinn 12:58, 30 May 2026 (UTC)[]

Large amount of edits at a rapid pace -- this and this are 1 minute apart, and there's more where that came from -- with summaries characteristic of current LLMs, and talk page comments that are also characteristic of AI. Some have been reverted, most seem to not have been. Gnomingstuff (talk) 03:59, 3 May 2026 (UTC)[]

There is so much AI slop, I'm going to have to look into how to handle this. Dr vulpes (Talk) 05:10, 3 May 2026 (UTC)[]

I would like to raise a process concern regarding the handling of this case. On June 4, Gnomingstuff reverted the HIPAA article to a "pre-AI version," removing all edits since March 2026. The edit summary stated "see user talkpage," but no message was left on my talk page or on the article talk page.

This revert removed content that had been proposed through the proper WP:COI talk page process and received independent approval from Compliance-Expert:

• The 42 CFR Part 2 section (approved March 12: "Perfect. Great job.")
• The HIPAA Security Rule NPRM section (approved March 16: "looks good.")

I also note that the original report filing this case cited "Large amount of edits at a rapid pace" as a concern. However, a review of Gnomingstuff's recent contributions shows a similar pattern: many of their recent reverts are less than 30 seconds apart across unrelated articles, raising questions about whether individual article content — including whether edits had gone through talk page review — was evaluated before reversion.

Per WP:BRD, I have raised this on the HIPAA article talk page and welcome community input. Jgellatly (talk) 22:25, 5 June 2026 (UTC)[]

Did you use AI? Yes or no. Gnomingstuff (talk) 22:28, 5 June 2026 (UTC)[]

My concern here is about process, not content origin, which was legitimate. Talk-page-approved content was mass-reverted without discussion, in violation of WP:BRD. I'd welcome input from other editors on whether that process was appropriate. Jgellatly (talk) 05:12, 13 June 2026 (UTC)[]

So, yes. Frankly most of the talk page comments appear to be straight-up LLM responses. As noted, having another low-edit COI account "approve" your LLM-generated material does not create a road block for removal. Sam Kuru (talk) 12:50, 13 June 2026 (UTC)[]

There's some question about the independence of user Compliance-Expert given that they admit to undisclosed conflict of interest on their Talk page, covering topics related to the articles being discussed here. See User talk:Compliance-Expert § March 2026. M kuhner (talk) 06:26, 13 June 2026 (UTC)[]

That truly is an interesting Talk page. I recommend that anyone tracking this case read the whole thing, not just the discussion of Gnomingstuff's reverts. M kuhner (talk) 07:00, 13 June 2026 (UTC)[]

@Admins willing to patrol AINB: I request a block of Jgellatly due to LLM use. Editor has been warned [1]. When asked Did you use AI? Yes or no. Jgellatly replied My concern here is about process, not content origin, which was legitimate.[2], a typical evasive LLM response. Their posts at HIPAA Talk appear mostly LLM generated (e.g. [3], [4]). They added the same pair of references, one malformed, to 30 different articles between 17:56 [5] and 18:09 [6] on 2026-04-03. Major edits have LLM-like summaries [7]. An edit request at § Proposed addition: HIPAA Security Rule compliance tools in Security Rule section had source integrity issues typical of LLM.[8]. I feel it's important to act now to prevent future cleanup issues, especially given the 30 edits in 13 minutes on 4/3. We have had no feedback from the editor that they will commit to non-use of LLMs in future, nor to help clean up the damage they've done. M kuhner (talk) 22:35, 15 June 2026 (UTC)[]

I wrote this, not AI. I have put a lot of work into the community, with no AI assistance. It was just ironic to be flagged for "AI" by someone who clearly does use AI. My mistakes: (1) batching a lot of issues over time and fixing them in an afternoon, and (2) using AI to find better links for us, a practice I have discontinued when it was inaccurate. We had a lot of misinformation on the pages in my field of expertise. If I'm not blocked, I'll continue to contribute. Jgellatly (talk) 23:10, 15 June 2026 (UTC)[]

I believe you when you say this comment was human-written. Thank you. We will see what the admins say. In the meantime, you might consider fixing the 30 broken references you introduced on 4/3. A few have been reverted by other editors, but most persist. Note that StatPearls is not a book so {{cite book}} is not appropriate, and the reference you give can't be parsed correctly as "work" means nothing in a cite book template. I think you introduced just that reference to 4 additional pages, and those are also broken. (Note that I am not making any claim this will help you: I'm not an admin and it's not up to me. But I, personally, would appreciate it. Otherwise I have to do it, and I'm not the one who made the mess.) M kuhner (talk) 23:38, 15 June 2026 (UTC)[]

De Insomniis (talk · contribs · deleted contribs · logs · filter log · block user · block log) has been active on the project for several years, but in recent months appears dramatically accelerated their output, apparently using LLMs to generate several articles. I've draftified a few (Draft:The_Descent_of_the_Dove:_A_Short_History_of_the_Holy_Spirit_in_the_Church, Draft:Hermitage of the Annunciation, Draft:Elliott House of Studies) and G15'ed ones that had hallucinated/nonexistent citations (Robert C. Tuck, War in Heaven (novel), The Mountain and the Valley (recreated by another user), Christ Church, Shelburne, George Westhaver). Another had LLM content stripped out by a reviewer (The Greater Trumps). There are others that don't immediately scream LLM or have fake sources, but the page creator's recent history gives me pause:

• Henry Ward Cunningham
• Assassination of Charles the Good
• The Lord of Spirits (podcast)
• Ricky Berwick
• Halifax Humanities Society
• King's College Chapel, Halifax

De Insomniis did not respond when asked about potential LLM use by Boynamedsue and has not edited since 27 April and thus has not responded to my warning on 1 May. Given that there are several articles possibly affected by undisclosed and unreviewed LLM output, I'd welcome insight from the experts on this board as to whether any of these need remediation. Dclemens1971 (talk) 13:33, 6 May 2026 (UTC)[]

User has returned to editing, but only to PROD an article. Will bear watching. M kuhner (talk) 06:50, 5 June 2026 (UTC)[]

THis user has persistently used AI to add content pertaining to the Kurukh language: [9] I have repeatedly warned them about WP policy regarding this, yet they continue doing this. 🪐Kepler-1229b | talk | contribs🪐 03:44, 10 May 2026 (UTC)[]

Agreed. Skidbot4 noticed and reverted on Kurukh language in late April, and content (including some markdown formatting) was reinstated and subsequently reverted at least once after that. Kurukh phonology was created by the user, and continues the pattern. Some other pages may also be affected.

They may be an element of content dispute here, note this talk page proposal which exhibits the usual format of LLM-written talk page text (WP:CEQ). Seercat3160 (talk) 04:52, 10 May 2026 (UTC)[]

I would like to add that the "proposal" is basically like "proposing" a new tense for English, it's nonsensical and the citation is hallucinated. Skidbot4 (talk) 07:54, 10 May 2026 (UTC)[]

Just FYI @Kepler-1229b, it usually isn’t best to tell the editor about WP:AISIGNS for WP:BEANS reasons, we don’t want them to still use LLMs but mask the ways we tell Kowal2701 (talk, contribs) 10:27, 10 May 2026 (UTC)[]

Thanks for letting me know. I'll take it into consideration next time. 🪐Kepler-1229b | talk | contribs🪐 17:01, 10 May 2026 (UTC)[]

The user has removed my AI notice in the talk page and is adding false content to the page again. Skidbot4 (talk) 06:01, 17 May 2026 (UTC)[]

The user continues to edit the page and has created another page https://en.wikipedia.org/wiki/Pronouns_in_Kurukh which still shows a few signs of AI. Something has to be done about this. Skidbot4 (talk) 17:31, 1 June 2026 (UTC)[]

• Michelle904 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This user was warned for potential AI use in 2025 and said they aren't using AI. However, I find this hard to believe, given the prose of their articles (very down-the-line typical of 2026-era LLMs specifically, and the fact that many of the edits are made in extremely quick succession, as in 3-4 minutes (example). The patterns of LLM output also track with the dates the edits were made -- for instance, their mid-2025 edits such as this are quite different in formatting, prose, etc, in the same ways that mid-2025 LLM output is different than the stuff we get today. Gnomingstuff (talk) 16:20, 13 May 2026 (UTC)[]

Thanks for your feedback.

I'd like to clarify a few things.. if I use AI, it's only to help with the writing style and all the facts are carefully researched and cited from real sources. About the time between edits, it's because I prefer to draft everything on my computer first, some of these sources are complex studies that take time to read and translate. It's a bit discouraging to be questioned when I'm putting in so much effort to get the work solid. But I'm still learning how things work around here and I really want to make sure I'm following all the rules. I'm open to your feedback to make sure my contributions are up to standard. Michelle904 (talk) 17:54, 22 May 2026 (UTC)[]

@Michelle904: Thank you for finally responding after concerns were raised by various editors here and on your talk page over many months. The most important thing that you need to know now is that using AI to help with your writing has been prohibited since March. That includes help with the writing style, so you must not do this in any future editing. Because AI makes it very easy to generate large amounts of text that cannot be trusted, it creates a disproportionate amount of work for human editors to review, as you can see by the slow progress being made on reviewing your edits listed below. —In solidarity with Wiki Workers United · ClaudineChionh (she/her · talk · email) 03:59, 23 May 2026 (UTC)[]

My goal was to save effort for editors.. not to make more for them. Thank you for making it clear and I will make sure my contributions follow the rules. Michelle904 (talk) 17:25, 23 May 2026 (UTC)[]

Added due to the vast majority of content on this article being AI generated. signed, A Random Hylian (Parry)(swing) 15:20, 14 June 2026 (UTC)[]

• PeppermintSA (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Extremely prolific AI use (disclosed as ChatGPT on their talkpage) going back a long time -- this user is singlehandedly responsible for "Sarajevo"-related tokens showing up disproportionately in my datasets, because of the sheer amount of their drafts that were rejected for LLM usage -- and still going to this day. They said they'd stopped using AI, but articles like this make that hard to believe. Gnomingstuff (talk) 16:24, 13 May 2026 (UTC)[]

Not sure why this is marked as resolved, this person has dozens of articles, not 1 Gnomingstuff (talk) 06:33, 15 May 2026 (UTC)[]

Yeah this and the previous editor have so many contributions that we have to dig through to even compile a list. ClaudineChionh (she/her · talk · email · global) 03:21, 19 May 2026 (UTC)[]

Oh, I see ... the "article cleanup list" comes from the beta version of the AINB Helper script. ClaudineChionh (she/her · talk · email · global) 03:58, 19 May 2026 (UTC)[]

ZS Khumalo (talk · contribs)
I didn't want to make this report as a lot of the articles they've created are sorely needed, but this user appears to still be using LLMs to generate content after having received two warnings (Dec 2025, Feb 2026); one of their drafts (Draft:Herbert Theledi) got rejected for AI use a few days ago. They wrote it in Feb 2026, when they were warned (courtesy ping Gurkubondinn); at that time they replaced a 404 link with a hallucinated source among other 'fixes' and subsequently removed the page's AI tag [10][11], while also denying AI use [12]. Unfortunately I can't access the 3rd edition of A History of Swaziland (1988) which they use a lot, but comparing some content to the 1972 ed., it looks like the usual "LLM getting creative" WP:V issues. In this April 2026 edit they add what appears to be a hallucinated source, which they've used in several articles. In this version of Chief of Tembe refs 4 and 5 don't support The Chief of Tembe oversees traditional governance within the Tembe Tribal Authority, manages land allocation, presides over customary courts and represents the Tembe community in provincial traditional leadership structures. in the slightest. Kowal2701 (talk, contribs) 20:14, 14 May 2026 (UTC)[]

I haven't pinged or notified them Kowal2701 (talk, contribs) 20:27, 14 May 2026 (UTC)[]

This user also created an article with chatbot communication:

This is just a starting point. We can expand and refine the article together! What would you like to add or change?
— Diff/1290554474 on Cambridge Makhosonke Dlamini

Then in Diff/1299120724, the user cited a book with the title "The History of Emaswati in South Africa" using the ISBN 9780620653848 in the reference, but that ISBN appears to be assigned to a different book ("Dark Clouds, God Has Been: A heartbreaking true story of a brave woman of God"). They also cited it in Diff/1338976350 on Mekemeke.

Tried asking on Talk:Cambridge Makhosonke Dlamini § LLM, but never got an answer. --Gurkubondinn 21:33, 14 May 2026 (UTC)[]

Deleted per G15. Chaotic Enby (talk · contribs) 16:42, 20 May 2026 (UTC)[]

Thanks for taking care of that. --gurkubondinn 17:18, 20 May 2026 (UTC)[]

Draft:Eswatini National Museum (which was written 18 May) has copyvios in the lead, cited to [13]. EARWIG

The article:

It is located in Lobamba, the country's traditional spiritual, cultural and political capital, just east of the Ezulwini Valley.

The museum was built in 1972 and serves as the headquarters of the Eswatini National Trust Commission (ENTC), the custodian of the nation's heritage and cultural archives

The source:

Eswatini’s (Swaziland’s) spiritual, cultural and political heart lies at Lobamba, just east of the Ezulwini Valley.

Eswatini’s national museum, known in siSwati as Umsamo Wesive was built in 1972 and sits just behind parliament. It serves as the HQ of the Eswatini National Trust Commission (ENTC), custodians of the nation’s heritage and cultural archives

@Admins willing to patrol AINB: , there's markup here 18 May Kowal2701 (talk, contribs) 14:43, 20 May 2026 (UTC)[]

There are phrases taken verbatim from that source peppered through the article. It makes me wonder what other copyvios there may be. ~Anachronist (who / me) (talk) 16:31, 20 May 2026 (UTC)[]

Not the most knowledgeable about dealing with copyvios, so I'll leave the judgement on this one to another admin. Given the repeated AI use, the copyvios, and the lack of constructive communication, I'm enclined to go for a partial block, although most of their contributions are already in draftspace, which may make it harder for them to prove good faith down the line if they're blocked from there. Chaotic Enby (talk · contribs) 16:45, 20 May 2026 (UTC)[]

A pblock from article space (to protect from more AI-generated edits) seems reasonable to me. --gurkubondinn 17:23, 20 May 2026 (UTC)[]

Yeah, in the block notice maybe be upfront re [14] having incontrovertible evidence of LLM use. If they don't respond well to that, there's unfortunately not much more we can do. I'm probably going to stubify some of their articles rather than PROD if it comes to that Kowal2701 (talk, contribs) 17:33, 20 May 2026 (UTC)[]

Good call, going to do that. Chaotic Enby (talk · contribs) 17:42, 20 May 2026 (UTC)[]

Responding to the ping. I second Chaotic Enby's article-space block of ZS Khumalo for LLM misuse. The most charitable reading of the situation is that the LLM generated the copyright violations, which means that the partial block should be enough for now. — Newslinger talk 17:54, 20 May 2026 (UTC)[]

Hi there, thanks a lot everyone. I wish to reply to the concern being raised. I only saw the discussions now. ZS Khumalo (talk) 13:02, 21 May 2026 (UTC)[]

Please do so. Fermiboson (talk) 14:24, 21 May 2026 (UTC)[]

I actually want to prepare a formal apology. I want to prepare the type of texts that will show my background and my daily enthusiasm in editing and writing on Wikipedia, mainly on Southern African politics and history - particularly about the royal history of Eswatini and the liberation history of South Africa.

Considering the accusations placed against me here, it actually hurts I must say and I actually must amend this broken internal relations/trust so that in the future eventually there is no misunderstanding about the tasks on our hand to make digitally the histories of people whose heritage is steeped in orals.

I have a duty to share all of my sources, including the A History of Swaziland (1988) 3rd edition screenshots page by page where issues of verifications/sourcing are raised as well as the full barcode of the book I bought in 2024 called History of Emaswati in South Africa.

With your permission, will it be possible that I get the email of some of you where -

1.) I can send texts regarding the current issue that led to me being painfully blocked

2.) Acknowledgement of the concerns raised by admins here and offer an unconditional apology, also seeking understandings on what not to be repeated in the future

3.) A background about my love for writing history

4.) Screenshots of books/sources I use to improve, expand or create articles on precolonial Eswatini history. ZS Khumalo (talk) 10:34, 29 May 2026 (UTC)[]

@ZS Khumalo: feel free to email me (you can't send attachments through this but I'll email you back so you can reply with the attachements). But I think that you should do nr. 2 onwiki. Take your time with it if you want though, and feel free to send me a draft of it via email first if you want to get my opinion on it or something, but I really think that you should say that onwiki rather than privately. --gurkubondinn 10:55, 29 May 2026 (UTC)[]

Thanks so much for your kind words. The type of device I use it gives me problem going through the Email Link.

If you don't have a problem at all, pls send me a 'Hello Zama, this is a Wiki admin regarding your issue' on zamastevenkhumalo👁 @
gmail.com 🙏🙏🙏

I'm sorry about this if it happens to inconvenience you.👌 ZS Khumalo (talk) 11:11, 29 May 2026 (UTC)[]

I got your email (I actually got two of them) and I've sent you an email back so you should have my email address now. I'm not an admin on Wikipedia though, just so you know. --gurkubondinn 11:25, 29 May 2026 (UTC)[]

Yes I got it. Thanks colleague. I confirm receipt. I will send the formal texts concerning my issues in 7 days. You will then advise me how to turn my statement of apology to the public discussion page (HERE) and policy understandings engaged with you, so that eventually we work in cooperation and we don't in the future find ourselves in the current issue, which I deeply regret.

Finally, it is therefore an obvious reality that the user of myself, who doesn't use a hidden name (pseudonym), has no interest in defeating the mission of the Wikimedia Movement or make its work disputed across by intellectuals and make its articles doubted by the world.

Ta ZS Khumalo (talk) 13:16, 29 May 2026 (UTC)[]

It is fine for you to collaborate with another editor on formulating an appeal, but keep in mind that Wikipedia business is best kept on Wikipedia, on publicly-accessible talk pages. No administrator is going to negotiate an unblock through email channels. If you send me an email, I will not reply to it, but I may reply on your talk page. ~Anachronist (who / me) (talk) 15:49, 29 May 2026 (UTC)[]

Besides the screenshots of books/sources, which shouldn't be uploaded on-wiki due to copyright reasons (roughly, for books from 1932 and later). Chaotic Enby (in solidarity · talk · contribs) 16:10, 29 May 2026 (UTC)[]

i don't want it to sound like a horse-trade agreement. In fact my charges are up publicly and is where I must answer. So certain screenshots won't happen here, otherwise I thought I would address myself to an admin, not an editor. This is so serious consciously because it has stopped my daily habits (reading and writing Wikipedia).

I will send an email, an email which must contain attached screenshots so that as I agree to my wrongness but then there is a general understanding about my zeal to document the heritage of Eswatini people. ZS Khumalo (talk) 19:37, 29 May 2026 (UTC)[]

@Anachronist and myself are both admins, if that is what you are looking for, and are telling you to provide your explanations on-wiki (besides images of sources containing copyrighted material) for the sake of transparency. Chaotic Enby (in solidarity · talk · contribs) 07:03, 30 May 2026 (UTC)[]

You are the main person that I should actually address the case.

In this particular case, you have a mandate to defend Wikipedia.

To me it appears correctly that I should address myself to you and is the reason said my issue should never through a horse-trade agreement.

Pls type me something on zamastevenkhumalo@gmail.com, where I can address verification concerns and where advice to turn our email conversation up here, so that it's clear what

1.) I'm charged with

2.) What are the crucial policies and mechanism to stick to them

3.) What are some arguments (because there are certain raised issues that are put and needs explanation in screenshot contexts, otherwise my case won't work because it needs sealed or existing evidence of my the books I use to expand, improve or create articles on precolonial Swazi history)

I actually need your e-mail too, so that my story is explained to you and we work towards an era of understandings (few words workshop) so that Wikipedia stays as it used to be traditionally without fake input. ZS Khumalo (talk) 20:33, 30 May 2026 (UTC)[]

Sorry, I haven't received my mandate to defend Wikipedia yet, I believe it should soon arrive in the mail with my baby globe plushie.For a very short recap: 1/ you've been partially blocked (from editing articles) for misusing AI to write articles; 2/ our relevant policy is this one, you might also want to read the guide to appealing blocks; 3/ it really depends on your extent of AI use, but being transparent about it and committing to a no-AI restriction in the future is a good start. Chaotic Enby (in solidarity · talk · contribs) 22:29, 30 May 2026 (UTC)[]

We don't conduct Wikipedia business over email, sorry. You have instructions on how to appeal the block. Your job now is to convince the community that unblocking you will be a net benefit to Wikipedia. ~Anachronist (who / me) (talk) 23:06, 30 May 2026 (UTC)[]

User was previously warned about their LLM use, which they responded to by admitting to use AI "in accordance with Wikipedia’s sourcing and neutrality guidelines" but without actually ceasing the use of AI tools. They have continued to create new pages showing clear AISIGNS. Other, non-page creation edits also show AI signs such as [15] and [16], some of which are in contentious topics. StartOkayStop (talk) 20:40, 18 May 2026 (UTC)[]

@StartOkayStop Hi, since the said previous warning - where I indeed responded and explained myself - my edits have been entirely researched, written, and reviewed by me. I'm doing my best to comply with Wikipedia’s sourcing, neutrality, and verifiability; especially with trying to stay within a "distant" and encyclopedic writing style (I try). If some specific case of sourcing, neutrality, copyright infringement or due-weight raise concern, I'll be happy to discuss and improve them on a case-by-case basis. Thank you for reviewing. Kaspar-trout (talk) 22:18, 18 May 2026 (UTC)[]

Thank you for your reply, but just to clarify, has 100% of the text added after 26 April been your own creation (no LLM involvement)? For example is the text in Special:Permalink/1354856263 entirely of your own construction? fifteen thousand two hundred twenty four (talk) 22:54, 18 May 2026 (UTC)[]

@Kaspar-trout: Could you please respond to @Fifteen thousand two hundred twenty four's question? It looks like you're continuing to make edits that are AI-generated, such as Special:Diff/1355018008. Please be aware of WP:NOLLM; as a guideline, using AI to create article content is prohibited. StartOkayStop (talk) 17:26, 19 May 2026 (UTC)[]

@Fifteen thousand two hundred twenty four and @StartOkayStop Hi both. No, I'm not. Please tell me what makes you think I'm using AI. We literally had an exchange yesterday, why would I use AI now. and what in that Norman mailer makes you think of LLM ebased adits. it's literally been written by me on a word document after reading archives from the NYT. and @Fifteen thousand two hundred twenty four I'd appreciate if you could give me precise indications of AI markers in those exemples you mentioned (which I doubt you can since i haven't make use of it). I don't mean being confrontational and I appreciate the work you do on this platform for making sure all the diting comply with Wikipedia's policies. Unfortunately I can't prove you anything just as much as you can't prove my writing is based on LLM use. Everything is based on good fate. I heard your warining, I answered in good fate. I read Mailer's books and articles, I know his boxing passion, I researched in newspapers and edited the page. I am happy to know exactly what seems wrong with my contribution and respond accordingly. Thank you both of you Kaspar-trout (talk) 20:19, 19 May 2026 (UTC)[]

I didn't state that AI markers are present, but I did ask if the text of Special:Permalink/1354856263 was entirely of your own construction, and this doesn't seem to have gotten a response. I'm especially curious about provenance of the contents of the "Philosophy and development model" section. fifteen thousand two hundred twenty four (talk) 20:54, 19 May 2026 (UTC)[]

Would be helpful to know why the statement The promotion described itself as a "digital-first platform" focused on storytelling and personality-building around fighters. is sourced to an article which does not contain the word "digital" (nor "storytelling", nor "personality-building", but at least those are not in quote marks). Putting the words aside, I don't see these concepts in the source either. Perhaps you could point them out to me? M kuhner (talk) 04:42, 20 May 2026 (UTC)[]

"Built as a digital-first platform, Usyk17 Promotions will emphasize storytelling, media presence, and strategic career development as its core competitive advantages." (1) (It's here in this ESPN article) and I just quoted it as is because it's a quote (I forgot to quotify it though). Thank you @M kuhner Kaspar-trout (talk) 07:30, 20 May 2026 (UTC)[]

The ESPN article would have been a fine source for this statement. But I was responding to Special:PermanentLink/1354856263 which sources it to [17]. It definitely is not in that source. So the article, as of that revision at least, falsely attributed a direct quote from ESPN to FanNation. M kuhner (talk) 15:48, 20 May 2026 (UTC)[]

Morning @Fifteen thousand two hundred twenty four I think sources arecitet to be read and crosse read when curious for more information. I decided to call the section "Philosophy and development model" hoping that this would be developed further in the future when more details on the promotion will be available and that the specific angle and positioning of Usyk promotional company would be more documented. Now we know that Usyk works with a high degree ofidealism, identity building, political positioning and so on (and we know he contributes with his businesses to the Ukrainian defence). I hope your question have now gotten an answer. Kaspar-trout (talk) 07:37, 20 May 2026 (UTC)[]

arecitet to be read and crosse = are cited to be read and cross (sorry for this) Kaspar-trout (talk) 07:39, 20 May 2026 (UTC)[]

Are you using Word's grammar check/rewriting features? Those are Copilot-based now. Gnomingstuff (talk) 23:50, 19 May 2026 (UTC)[]

@Gnomingstuff, I have grammarly installed Kaspar-trout (talk) 07:38, 20 May 2026 (UTC)[]

Got it -- that's what's causing this, since Grammarly is AI-based.

That being said, if Grammarly (or any similar tool) is changing the text enough that someone else can identify the text as AI-assisted, then that's going beyond the "basic copyediting" allowed in WP:NEWLLM. Gnomingstuff (talk) 11:10, 20 May 2026 (UTC)[]

@Gnomingstuff thank you. Good to know, but I don't believe it changes the text so much as i verify it constantly. Even today I reviewed some of these articles that have been "identified" and they still had some syntax issues, awkward grammar and little imperfections that i'm trying to correct. I really do write this material and read my sources. Kaspar-trout (talk) 11:17, 20 May 2026 (UTC)[]

I think this might be a misunderstanding of the guidelines. The "basic copyediting" allowed under WP:NEWLLM is stuff like spelling, capitalization, or punctuation, where there is no way anyone would be able to tell whether "beleive" was turned into "believe" by AI, by a spell-checker from 1993, or by a paper dictionary.

Whereas something like These publications have been identified by critics and historians as among the most influential examples of artist-run magazines, reflecting successive shifts from avant-garde experimentation and conceptual art toward postmodern theory, counterculture and alternative editorial infrastructures. is not only identifiable as AI-assisted, but has source-to-text integrity issues, given that it is cited to a blog post that doesn't contain the words "postmodern" or "post-modern," or even just "modern"; doesn't state or imply that these are successive shifts given that many of these magazines seem to have operated more or less at the same time and there's not really a chronological "shifts" argument being made by the post; and is the opinion of one blogger, not multiple "critics and historians." So either this is a change of meaning introduced by AI (since these are the kinds of problems that AI introduces all the time), or you didn't read the source very carefully. Gnomingstuff (talk) 12:25, 20 May 2026 (UTC)[]

I added post modern because I know what post modernism is and know how to contextualize it, since we also need to be careful to copyright infringements and need to find ways to edit and write encyclopedic information without copying our sources. Kaspar-trout (talk) 12:30, 20 May 2026 (UTC)[]

That's not how Wikipedia works. Per the verifiability policy: Even if you are sure something is true, it must have been published in a reliable source before you can add it. So if a source doesn't call something postmodern, then you can't use that source as a citation for it being postmodern. The no original research/synthesis policy is also relevant here. Gnomingstuff (talk) 12:40, 20 May 2026 (UTC)[]

I understand what you say, but what i've added here it's just sensible literate cultural knowledge, using words that allows me to summerize without infringing copyright and staying truthful. I am happy to rephrase if necessary and stick to the source or find more sources if this will help. Nevertheless, I think this conversation helped understanding the issues initialy addressed. I'll bonify this paragraph. Thank you for your help @Gnomingstuff Kaspar-trout (talk) 12:50, 20 May 2026 (UTC)[]

I have further concerns regarding this editor's potential use of AI.

• First is this edit to 2026 Strait of Hormuz crisis, which is cited to this NYT's article despite it never mentioning "global energy transit", the time-span of "decades", "global supply chains", nor any "international debate[s]".

I had initially assumed this was an issue of WP:SYNTH &/or WP:OR, however...

• This edit to Israeli–Palestinian conflict has several citations, however their given page numbers seem completely unrelated to what they're theoretically meant to support

• Rashid Khalidi's The Iron Cage: The Story of the Palestinian Struggle for Statehood doesn't even mention "Arabism", let alone "Pan-Arabism" within any of the given page ranges 18–33, 31–45, 145–154.
• Michael Doran's Pan-Arabism Before Nasser: Egyptian Power Politics and the Palestine Question only uses the term "Palestinian" in one of the four page numbers, but not in a way that supports the article content.
• Rashid Khalidi's book is cited again with no mention of "Zionism" within the pages 145–177.
• The 2nd edition of The War for Palestine: Rewriting the History of 1948 as edited by Eugene L. Rogan & Avi Shlaim never refers to a "charter" & only refers to the "West Bank" once, but again, not in a way that supports the article content.
• The Avalon Project's documentation of The Palestinian National Charter lists article 24 as "The Palestinian people believe in the principles of justice, freedom, sovereignty, self-determination, human dignity, and in the right of all peoples to exercise them.", not “This Organization (PLO) does not exercise any regional sovereignty over the West Bank in the Hashemite Kingdom of Jordan, on the Gaza Strip or in the Himmah Area.”, with the "West Bank" never mentioned. This is due to Kaspar-trout referring to the 1964 charter, but cited the revised version from 1968.

• The quote is still wrong however as the initial charter from 1964 stated "This Organization does not exercise any territorial sovereignty over the West Bank in the Hashemite Kingdom of Jordan, on the Gaza Strip or in the Himmah Area." (Emphasis mine).

I was originally going to bring this up at their talk page, but this seems to be a more appropriate venue - Butterscotch Beluga (talk) 19:33, 9 June 2026 (UTC)[]

@Butterscotch Beluga Hi, so just for your information here's what I have right here right now, taken from the Iron cage book, and I'm not going to do this work for you for all the citations you just mentioned:

"...American and allied policymakers supported them against such identified enemy forces as Arab nationalism, Pan-Arabism, local communist parties, radical regimes, Palestinian nationalism, and later the Soviets in Afghanistan​" (p17-18)

"In his autobiography, al-‘Isa relates one of the critical moments in Damascus when the balance between his loyalty to Pan-Arabism and his Palestinian identity was most sorely tested." (127)

"Loyal though al-‘Isa was to the ideology of Arabism and to Faysal, his higher loyalty was clearly to Palestine. He resigned immediately after this episode, but Faysal reinstated him after his return from Europe, and he remained chief secretary of the Royal Diwan until the fall of Faysal’s government in July 1920." (p.128)

"Like all matters involving language and ethnicity in this era, this rapidly became a “national” issue, with a heightened awareness of their Arabism among the Orthodox Arabs of Syria and Palestine as one of the primary consequences." (p.130)

"For more details on the Palestinian press see Rashid Khalidi, Palestinian Identity and also “Anti-Zionism, Arabism and Palestinian Identity: ‘Isa al-‘Isa and Filastin,” in Political Identity in the Arab East in the 20th Century, ed. Samir Seikaly (Beirut: American University of Beirut Press, forthcoming)." Bibliography (p.245)

In the index you can even find all the Pan-Arabism notes....

This is for The Iron Cage, which I have and read. Thank you Kaspar-trout (talk) 20:18, 9 June 2026 (UTC)[]

To address your concern directly: I have sometimes used ChatGPT as an aid in locating sources, but not as a source and not as a substitute for verification. I review the underlying material myself before adding content, and I am willing to correct citations where edition differences or page-number discrepancies create confusion. As I said before, I usually gather notes from source material, verify it, and summarize it concisely in my own words, while also taking care to avoid close paraphrasing or copyright issues (which isn't always easy). I understand the concern about machine-generated or poorly verified edits. That is not how I am editing. Kaspar-trout (talk) 20:49, 9 June 2026 (UTC)[]

also: The Palestinian National Charter Article 24: This Organization does not exercise any territorial sovereignty over the West Bank in the Hashemite Kingdom of Jordan, on the Gaza Strip or in the Himmah Area. Its activities will be on the national popular level in the liberational, organizational, political and financial fields. (I have different documents from may 28 1964, with this article mentioning the West Bank - but this is another subject matter) Kaspar-trout (talk) 20:57, 9 June 2026 (UTC)[]

I am very confused.

ButterscotchBeluga says that the edit cites pages 18-33, 31-45, and 145-154 of The Iron Cage.

In your message just above you say that the material is on pages 17-18, 127, 128, 130, and 245.

These do not match. It cannot be a book version problem: you put those citations into the article, and you wrote the text above. You presumably have a physical copy of the book. ButterscotchBeluga could have been looking at a different version when they made their comments, but that would not explain why the article and your message above don't match.

Why don't they match? Is it because your LLM assisted workflow makes up page numbers?

It is, as you clearly have found, a lot of work to check and verify these citations. That's why the LLM workflow is bad. It's more work to fix inaccurate LLM text than to write human text which is accurate in the first place.

M kuhner (talk) 21:14, 9 June 2026 (UTC)[]

First quote (p XXII) - List of those the US identified as enemies in the Middle East. This doesn't support characterizing these movements as "intertwined" with the "Palestinian political identity". Also the source says "mid-1950s until the early 1990s." not the "early 1960s"

Second quote (p 97)

Third quote (p 98)

Fourth quote (p 100)

Fifth quote (p 232) - I hope you understand why you can't cite a note telling you "For more details" read a different document. Especially as that note makes no reference to "a distinct Palestinian state".

We can assume the page discrepancy is due to different versions of the book as your given page numbers are mostly 30 ahead. However even if we dismissed this discrepancy & consider your numbers accurate, the page numbers you've given here still don't corelate with those you initially cited in the article. Butterscotch Beluga (talk) 21:20, 9 June 2026 (UTC)[]

Thank you for checking the sourcing. I agree that the previous version was confusingly cited, and that some of the wording was broader than the cited page ranges could comfortably support. I have revised the passage accordingly.

The new version removes the causal implication that Nasserism or Pan-Arabism “explained” the absence of a Palestinian state. Instead, it separates the claims more narrowly: first, that no independent Palestinian state was established in the West Bank or Gaza between 1948 and 1967; second, that Khalidi describes the post-1948 period as one in which Palestinians lost control over their fate and were “dispersed, divided, and caught between” Israel and the Arab states, which controlled the territory of former Mandatory Palestine; third, that Jordan and Egypt restricted independent Palestinian political activity in the West Bank and Gaza; and fourth, that Khalidi describes the Arab League’s creation of the PLO in 1964 as partly intended to contain and control growing independent Palestinian nationalism.

I also corrected the charter issue: the Avalon Project gives the revised 1968 charter, not the original 1964 text, so I have separated the 1964 Article 24 statement from the 1968 revised charter and corrected the wording to “territorial sovereignty.” The aim of the revision is to avoid synthesis and make each claim more directly tied to the cited source. Again, thank you for checking this edits and pushing me back into my notes and sources. I hope this addresses the previous issues. Kaspar-trout (talk) 08:34, 10 June 2026 (UTC)[]

This still doesn't address the concern I & M kuhner raised. If, as you say, you didn't use AI to write this material, then how did you arrive at the initial page ranges (18-33, 31-45, & 145-154)? They neither correspond to the page numbers I cited (XXII, 97, 98, 100, & 232), nor even those you cited in this discussion (17-18, 127, 128, 130, & 245). Butterscotch Beluga (talk) 17:00, 10 June 2026 (UTC)[]

@Butterscotch Beluga I understand why the discrepancy raises concern and after a more thorough look into my notes and sources, I should answer it directly.

The page-range problem came from my own note-taking and citation-control error. I was working from more than one version of the source: I have a French physical edition of the book, and I also had access to an English PDF. The PDF I was using appeared to use document pagination rather than publication pagination, which did not hit me immediately, and I mistakenly took notes using the PDF/document page positions rather than the published page numbers. When I later answered here, I checked against the physical book / proper pagination, which is why the page numbers I gave in the discussion no longer matched the ranges I had originally placed in the article.

That is not good citation practice, and I fully accept responsibility for it. I treated working notes as if they were ready article citations, and that was the mistake. I am not defending that practice. As I said above, I have used ChatGPT at times as an aid to find material to read, and sometimes to check English syntax or formulation when a sentence becomes confusing, since English is not my native language. But I do not use it as a source, and I should not have allowed any material into an article without a final source-by-source and page-by-page verification.

That is why I revised the passage, removed the broader causal framing, corrected the charter issue, and tried to make each claim more narrowly tied to a directly checked source. Going forward, I will not add material from mixed PDF/book notes unless I have first carefully checked the exact publication pages. My apologies for the confusion and for the extra work this created. Kaspar-trout (talk) 19:20, 11 June 2026 (UTC)[]

I tagged for LLM use (AILIST, nonexistent category, structure) and draftified; they responded by rewriting the article to comply with WP:NOLLM removing the list and the tag and moving it back. I₂Overcome ^talk 00:14, 20 May 2026 (UTC)[]

They re-recreated Blinkfire, now at AFD. fifteen thousand two hundred twenty four (talk) 10:31, 16 June 2026 (UTC)[]

Bmgupta12345 (talk · contribs · deleted contribs · blacklist hits · Edit filter search · what links to user page · count · COIBot · Spamcheck · blacklist hits · user page logs · x-wiki · status · Google · StopForumSpam)

Some stuff to clean up here. If anyone else could contribute it would be great. Thanks, CommonsKiwi (talk) 02:39, 20 May 2026 (UTC)[]

Are all of their edits LLM-assisted, or did this start after a certain date? A little bit more info about this could help. Sarsenet•he/they•(talk) 02:44, 20 May 2026 (UTC)[]

Not sure, they seemed to start by doing a lot of small fixes/copyediting. From what I saw these edits looked fine, even if they were AI-assisted. What concerns me is where they have added content, which has become more frequent over time. CommonsKiwi (talk) 02:54, 20 May 2026 (UTC)[]

The small edits are absolutely AI generated (note how the edit summaries get cut off, suggesting they didn't even bother to check what they pasted in). And from the looks of it, it is yet another deluge of Newcomer Tasks-induced AI editing. One day WMF will stop destroying their product with enshittification features. Gnomingstuff (talk) 03:54, 20 May 2026 (UTC)[]

User page also looks like it was done by ChatGPT. Sarsenet•he/they•(talk) 06:12, 20 May 2026 (UTC)[]

I marked 3 of this user's articles for G15 due to obviously implausible references. If I was to hazard a guess the rest of their contributions have the same problems. ~ A412 ^talk! 07:03, 20 May 2026 (UTC)[]

Tagged two more. Carol Nantongo and Chance Nalubega I can't find an obvious reason to G15. ~ A412 ^talk! 16:32, 20 May 2026 (UTC)[]

Tagged one more. Erick Kambale is weird. ~ A412 ^talk! 16:38, 20 May 2026 (UTC)[]

Another round of updates, now 7 G15s. I think Carol Nantongo and Chance Nalubega may be okay, they cite real references and score low on GPTZero. Not sure about Erick Kambale. ~ A412 ^talk! 18:53, 20 May 2026 (UTC)[]

Erick Kambale would probably survive AFD. User was blocked by Drmies. I'm realizing we probably need to look at their noncreations, I think I've added all post-2023 major contributions above. ~ A412 ^talk! 20:55, 20 May 2026 (UTC)[]

I looked at New Vision. I think it would survive AfD even if everything unsourced or dead-linked was removed, so I will do that. Won't be much of an article but at least it will be sourced. There's an interesting story here, hopefully someone can flesh it out further. M kuhner (talk) 18:46, 22 May 2026 (UTC)[]

Could use advice on New Vision: what to do with sources that sound plausible but are not accessible by me? The link to the Ugandan stock exchange never resolves, and a really interesting paper is paywalled except for the first few paragraphs (which suggest, but don't prove, that the information is actually there). Sourcing in this article was so bad I can't take anything on faith. M kuhner (talk) 04:30, 24 May 2026 (UTC)[]

Which reference is the interesting paper? ~ A412 ^talk! 05:43, 24 May 2026 (UTC)[]

Kirumira, Mark; Ajwang, Jan (3 May 2007). "Uganda: The Limping Newspaper Industry". Daily Monitor via AllAfrica.com. Retrieved 24 September 2017.

This one. M kuhner (talk) 05:52, 24 May 2026 (UTC)[]

Seemingly AI-generated revision to Bocconi University School of Law; uses primary sources mainly and references its own Wikipedia page. Subsequent discussion on user talk page with possible sock/meatpuppet seems to indicate that they may be using LLM-gen text to respond to comments as well. It’s very late where I am at the moment so I don’t have the energy to deal with this. If someone who’s skilled in dealing with things along these lines can step in, I’d appreciate it. OrbitalVoid49 (talk) 11:25, 20 May 2026 (UTC)[]

I inform you that there is no AI response system, I believe that these accusations are falling into defamation. I implemented a page that lacked a series of information by putting clear, external sources (such as the University website) in addition to other sources. I believe that on your part there is only the will to mock the work of others instead of improving Wikipedia.

In addition, I work with my name and surname not with a false name aimed only at protecting yourself ~2026-30092-25 (talk) 11:29, 20 May 2026 (UTC)[]

So is the ~2026 your surname, or is it the 30092-25?

I don't know what an "AI response system" is but that's not what they're asking. They are asking whether you (assuming you just aren't logged in currently) used AI to write the article, which is a question that has been dodged over and over here. Gnomingstuff (talk) 12:31, 20 May 2026 (UTC)[]

Even if it's untrue, "you used AI for editing Wikipedia" is not libel. Hint: more people sensed that you have used AI. But since it is not a crime, nor a reason for social dishonor, it does not amount to libel. I can say "you did not brush your teeth today". It is a lie, but not libel. tgeorgescu (talk) 22:17, 8 June 2026 (UTC)[]

Disruption continues. More eyes needed. tgeorgescu (talk) 19:26, 10 June 2026 (UTC)[]

Considering the article got turned into a redirect and the user itself got blocked from editing the page, should this be considered closed? OrbitalVoid49 (talk) 21:31, 13 June 2026 (UTC)[]

This user was already warned on May 14 to stop creating articles using LLM tools. Instead of responding to the message, they have continued creating AI-generated articles. Two days later, they created Peace Effiong, which includes the sentence: She has been listed among the attacking players representing Rivers Angels in youth competitions and development tournaments connected to Nigeria’s national team scouting system Just today they created Yusuf Abdullahi Tsamage, which says He has been described in match reports as an attacking player deployed in advanced positions This is clearly not verified by the attached source. JTtheOG (talk) 03:58, 21 May 2026 (UTC)[]

• Draft:Philippine House Quad Committee hearings (edit | talk | history | links | watch | logs)
• Rizalninoynapoleon (talk · contribs)

Found this page in article space, where it had been created with 6x OAICITEs: Diff/1355552206. I've draftified the article for now. Decided against tagging it for G15 deletion because the user who created it has the WP:AUTOPATROLLED user right, and I think that a closer look on their recent articles is warranted. By my count they have created 15 articles since November 2023, which should probably all be looked at. --gurkubondinn 17:15, 22 May 2026 (UTC)[]

I suspect that for example this edit was generated: Diff/1353594715. An experienced user like this would know to not use markdown syntax for |url= values in citation templates, but existing references were edited in this bizarre manner. --gurkubondinn 17:42, 22 May 2026 (UTC)[]

To whom it may concern,

Please note that the article remains a work in progress and is not yet ready for publication. Significant revisions, source verification, and cleanup are still required.

No AI tools were used in the drafting of the article. Its publication and subsequent editing occurred without my authorisation through actions taken by a former assistant who had access to the account. The individual is no longer employed and is no longer involved with the project. I apologise for any confusion or inconvenience that this may have caused. Rizalninoynapoleon (talk) 08:55, 26 May 2026 (UTC)[]

@Rizalninoynapoleon, articles that are not yet ready for publication should not be created in mainspace. I will be removing your autopatrolled flag for now. In solidarity, asilvering (talk) 15:37, 26 May 2026 (UTC)[]

Thank you for bringing this to my attention, @Asilvering

My apologies for the inconvenience caused. The article in question was an unfinished draft and should not have been created in mainspace in its present state. I understand the concern regarding the publication of material that had not yet undergone the necessary cleanup and review.

I have since reviewed and updated access to my Wikipedia account and implemented additional safeguards to ensure that only authorised edits are made going forward. I am also reviewing my draft workflow to prevent similar issues from occurring in the future.

I appreciate your action and understanding, and I accept the temporary removal of the autopatrolled flag. Thank you for your continued work in maintaining editorial standards across the project. Rizalninoynapoleon (talk) 15:45, 26 May 2026 (UTC)[]

The edit Diff/1357524000 added a new reference to an article, but it returned 404 not found and has been reverted. The user has also responded to a PROD notice on their talk page with an AI-generated reply. @Rizalninoynapoleon, could you please stop using AI here on English Wikipedia? --gurkubondinn 09:44, 3 June 2026 (UTC)[]

Unfortunately, they seem to have continued using AI, both for articles and communication:

• Special Assistant to the President for Investment and Economic Affairs: created with trailing markdown-style code block syntax (triple backticks, ```) in December 2023, so AI-use has most likely been going on for a while.
  • Too old for draftification, has been tagged for PROD in revision 1357424489. Has also been tagged as AI-generated.
• Draft:Philippine House Quad Committee hearings: created with 6x OAICITES, in article space. Blamed on a former assistant who had access to the account above.
  • Draftified
• Second impeachment of Sara Duterte: the user has made several edits to this article, but I have not combed through all of them. I've tagged the article as AI-generated, and commented in the merge discussion that is currently ongoing at WP:AfD/2026 Philippine Senate walkout.
  • Revision 1353594715: added references with Markdown-formatted links in the url parameters, and also formatted the url parameter in already-present references as Markdown. The edit inserted fictional/hallucinated references; several "soft 404s" that redirect to (or show) the main page ([18], [19], [20], [21], [22], [23], [24]) and two return 404 not found ([25]], [26]).
  • Revision 1353586365: added one hallucinated/fictional reference ([27])
• Flood control projects scandal in the Philippines: Tagged the § Bureau of Immigration section and refs, have not cleaned up yet, still looking through the rest of their edits to this article.
  • Revision 1357346300: added four hallucinated/fabricated references (404 not found: [28], [29], [30]; returns the main page of the site: [31]).
• Surrender of Jinggoy Estrada (a BLP): User has made other edits to the article this year, that will need to be checked.
  • Revision 1357524000: added an hallucinated/fabricated reference that returned 404 not found ([32]).Reverted in revision 1357556626.
• Tony Meloto (a BLP)
  • Revision 1357247860: added two fictional/hallucinated references that have never saved by archive.org. Both are "soft 404s" that show the main page of the site ([33], [34]).Reverted in revision 1357881427.
• Manuel Bonoan (BLP)
  • Revision 1357241059: reformats url of an already present reference as MarkdownReverted in revision 1357892876.
• Marcos Collection: created with fictional or hallucinated references, with eight in the current revision. Requested G15.

Communication:

• User talk:Rizalninoynapoleon § Proposed deletion of Special Assistant to the President for Investment and Economic Affairs: posts an AI-generated objection to a PROD tag, ignores my question and repeats the PROD objection.
  • § May 2026: ignored my questions, but eventually (and to their credit) replied in what is probably their own words. Claims in revision 1357871051 that articles where fixed with spell check, not AI.
• Talk:Special Assistant to the President for Investment and Economic Affairs § Refuting the proposed deletion: same PROD objection.

--gurkubondinn 06:19, 5 June 2026 (UTC)[]

AI aside, it seems concerning that this appears to be a WP:Shared account with an internal process for producing authorised edits? CMD (talk) 08:50, 5 June 2026 (UTC)[]

Yep.. I'm working on a tracking page... and it's a lot. --gurkubondinn 08:58, 5 June 2026 (UTC)[]

Where's the evidence of a shared account? In solidarity, asilvering (talk) 09:27, 5 June 2026 (UTC)[]

...editing occurred without my authorisation through actions taken by a former assistant who had access to the account.
— User:Rizalninoynapoleon in Diff/1356203228

They didn't say that the account was intentionally shared, but that someone else had access to their account. Which isn't the same thing. --gurkubondinn 09:35, 5 June 2026 (UTC)[]

They then said they "updated access to my Wikipedia account and implemented additional safeguards to ensure that only authorised edits are made going forward". A personal account doesn't require an authorisation to approve edits, nor safeguards to make sure others with access don't make edits that are not authorised. CMD (talk) 10:53, 5 June 2026 (UTC)[]

Could also be a WP:BRO variation. --gurkubondinn 10:57, 5 June 2026 (UTC)[]

Not saying that this is the case btw, but the part about "ensuring that only authorized edits are made" part sounds odd. I've never had to make sure that my account isn't making unauthorized edits. --gurkubondinn 11:26, 5 June 2026 (UTC)[]

I've definitely heard that phrasing spit out by AI before in regards to "lying about getting hacked and backpedaling" unblock appeals. Sarsenet•he/they•(talk) 11:45, 5 June 2026 (UTC)[]

Tracking page created:

• Wikipedia:AI noticeboard/2026-06-05 Rizalninoynapoleon

--gurkubondinn 09:30, 5 June 2026 (UTC)[]

Requested G15 for Daytoy ni Bannawag because it consisted almost entirely of hallucinated references, only one reference actually works. The current revision had the same set of references as when it was created (no additions or removals). Some of the same references had also been added to Ferdinand Marcos's cult of personality, reverted in revision 1358515157. One of the references was just a low-resolution photograph. ‑‑gurkubondinn 04:56, 9 June 2026 (UTC)[]

I don't understand what this is and have to run so don't have time to figure it out but maybe someone at this noticeboard can take a look at https://en.wikipedia.org/wiki/Special:Contributions/Wendy_at_AgenticCommons . (Sorry for this very lame post, will check back in tomorrow hopefully.) --JBL (talk) 18:04, 22 May 2026 (UTC)[]

It's an AI agent, @Admins willing to patrol AINB: Kowal2701 (talk, contribs) 18:10, 22 May 2026 (UTC)[]

👁 Red X
Blocked Wendy at AgenticCommons (talk · contribs) indefinitely, as operating an AI agent without prior approval is a violation of the bot approval policy.​ — Newslinger talk 18:20, 22 May 2026 (UTC)[]

I have asked Wendy at AgenticCommons some questions on their user talk page to give them a chance to explain their editing process. — Newslinger talk 18:44, 22 May 2026 (UTC)[]

I have unblocked Wendy at AgenticCommons after they explained their editing process (which includes manually submitting edits) and agreed to seek community approval before continuing their project. I recommended Wendy at AgenticCommons continue the discussion either here or at WT:LLM to ask whether their project is permissible under the WP:LLM guideline. — Newslinger talk 07:17, 25 May 2026 (UTC)[]

Thanks Newslinger for the unblock and for pointing me to this thread!

I'd like to clarify that every edit on this account is manually submitted by me after I review it. Here's the full workflow (also on my user page):

1) An automated process fetches Wikipedia images that are missing alt text.

2) An AI model (claude-sonnet-4.6-vertex) generates candidate alt text for each image.

3) I review each candidate and publish it manually under this account.

Also, I've already made the following changes to comply with the requirements:

1) Renamed my account from AgenticCommons to Wendy at AgenticCommons to comply with WP:ORGNAME

2) Created a user page to explain what Agentic Commons is, what we do, how we work and what my role is.

3) Removed the promotional parts in the edit summary and added the paid disclosure.

I hope this addresses the concerns, and I won't add any alt text until I hear back.

Thanks,

Wendy Wendy at AgenticCommons (talk) 08:39, 25 May 2026 (UTC)[]

One thing I’ll say is that we do desperately need more alt-texts (obv for WP:ACCESSIBILITY reasons), only a really tiny amount of images have it. Though I haven’t done a survey on whether the edits comply w WP:ALTTEXT. What prompts are you using? Kowal2701 (talk, contribs) 08:46, 25 May 2026 (UTC)[]

Hi Kowal2701,

Thanks for pointing out how much this matters for Wikipedia! The prompt was AI-drafted and we keep tuning it from real outputs for better results — below is the current version:

-----------

Prompt:

1. 1. Role

You are writing alt text for a Wikipedia image. Your audience is readers using screen readers who cannot see the image. Your alt text is what they hear in place of the image, read just *before* the caption.

1. 1. Task

Write the alt text string for the image below, following WP:ALTTEXT. Your output will be inserted as the |alt= parameter of the image's [[File:...]] tag and reviewed before publication.

1. 1. Guidelines (WP:ALTTEXT)

1. **Context drives the description.** Describe the image in terms of why this article uses it. The same portrait warrants different alt text in a biography vs. in an article about the painting itself. Read the surrounding context below before writing.

2. **Don't duplicate the caption.** Screen readers read alt text and then the caption back-to-back. If the caption already identifies the subject, the alt text should add what's *visually* informative (setting, composition, expression), not repeat the same sentence. When the caption fully describes the image, you may output exactly: `Refer to caption`.

3. **Front-load the essentials.** Put the key information in the first few words so screen reader users can skip ahead once they've grasped it.

4. **Be concise.** One short sentence is usually enough. Longer only for charts, diagrams, or maps that carry detailed information.

5. **Plain text only.** No wiki markup, no HTML, no line breaks, no citations.

6. **Don't start with "Photograph of…" or "Image of…".** Readers already know it's an image.

7. **Don't describe irrelevant attributes.** For Elizabeth II in a non-fashion article, write her name — not "an elderly woman in a black hat." For charts, the *information* matters, not the color or position of bars.

8. **Don't describe what isn't visible in the image.** Stick to what a sighted reader would actually see.

9. **Decorative or purely aesthetic images:** output one of `photograph`, `painting`, `sculpture`, `icon`, etc. — a minimal alt is better than blank for linked images (blank causes screen readers to read the filename aloud).

1. 1. Image

- Filename: `{filename}`

- Article: [[{title}]] — {article_url}

- Section: {section_title}

- Existing caption: {caption}

- View the image: {image_url}

- Image description page: {commons_url}

1. 1. Surrounding article context

The wikitext around the image, so you can see how the article actually uses it:

{context}

1. 1. Output

The alt text string itself. Plain text, one line. No quotes, prefix, explanation, or markup. It will be inserted verbatim into the `|alt=` parameter.

---------

Notes on what gets filled in:

1) {filename}, {title}, {section_title}, {caption} — pulled straight from the article's wikitext for that image

2) {image_url} — direct URL to the image file, so the agent actually views the image

3) {context} — ~500 characters of wikitext on each side of the [[File:…]] tag, so the agent can see how the article actually uses the image and craft the alt text based on the context

-----------

Thanks again for raising it — happy to keep talking if you any other questions!

Best,

Wendy

Wendy at AgenticCommons (talk) 11:10, 25 May 2026 (UTC)[]

Quick check-in — alt-text edits are still paused on my end. Any remaining concerns, or are we good to start back up? Thanks. Wendy at AgenticCommons (talk) 03:36, 26 May 2026 (UTC)[]

Do not start back up. The use of LLMs to generate or rewrite article content is prohibited per WP:NOLLM, there is no exception for alt-text, nor for reviewed content. fifteen thousand two hundred twenty four (talk) 03:46, 26 May 2026 (UTC)[]

I agree with 15224: There is no carve-out for alt text in our policy. Furthermore, using LLMs to generate alt text is controversial and not universally welcomed by accessibility advocates. —In solidarity with Wiki Workers United · ClaudineChionh (she/her · talk · email) 03:52, 26 May 2026 (UTC)[]

Pinging Moxy as they do work on this Kowal2701 (talk, contribs) 05:05, 26 May 2026 (UTC)[]

I agree with both 15224 and Claudine here. There is no exception for alt-text, and I don't think that our readers who rely on accessibility tools are any less deserving of a human-written encyclopaedia. --gurkubondinn 07:04, 26 May 2026 (UTC)[]

I’ve notified WP:ALTTEXT, WP:ACCESSIBILITY, and WP:WPACCESS Kowal2701 (talk, contribs) 17:57, 26 May 2026 (UTC)[]

Using LLMs to generate alt text is an insult to our readers who depend upon alt text. Stepwise Continuous Dysfunction (talk) 22:35, 26 May 2026 (UTC)[]

Thanks everyone who followed up on this. --JBL (talk) 20:05, 31 May 2026 (UTC)[]

They have many Wikimedia projects according to https://agentic-commons.org/projects. Secretlondon (talk) 19:09, 4 June 2026 (UTC)[]

(I did not include articles already deleted.) M kuhner (talk) 04:07, 24 May 2026 (UTC)[]

I've added the ones that have already been removed from mainspace, either via speedy deletion under G15 or via draftification. --gurkubondinn 09:31, 24 May 2026 (UTC)[]

• DddsDD (talk · contribs)

This user has been generating articles at an alarmingly high rate, I have just finished G15'ing and draftifying the nine articles that they have created today in under 90 minutes. There are more articles that need to be checked and cleaned up, but I can't get through more right now so I am posting here. They've created 57 articles in total, and a bunch of templates and categories as well. Common sense tells me that their edits to existing articles probably also need to be cleaned up as well. --gurkubondinn 17:28, 23 May 2026 (UTC)[]

Most of their articles have references with access dates from years ago, showing that the editor is citing sources that they haven't read. --gurkubondinn 17:31, 23 May 2026 (UTC)[]

It seems like the outdated access-dates are due to the articles being unattributed translations from other Wikipedias. In such cases, DddsDD still needs to follow the guideline Wikipedia:LLM-assisted translation, however. I did decline the G15 for Eldar Dzharakhov for this reason, but WP:LLMPROD could be an alternative, policy-compliant approach. Ca ^{talk to me!} 07:38, 25 May 2026 (UTC)[]

What? It doesn't matter if something is a machine translation or not, this is still machine generated text and it meets the G15 criteria of nonsensical references and this should not have been declined. There are access dates from over a decade ago! Several of the sources go nowhere. --gurkubondinn 08:55, 25 May 2026 (UTC)[]

Eh, it's a moo point now, I've stubified the article instead. --gurkubondinn 10:30, 25 May 2026 (UTC)[]

The dated access dates are due to the fact the original Russian Wikipedia article was created in 2017.[35] All the cited sources used to exist at some point, but simply underwent link rot given most are archived in the Wayback Machine. As the citations were not a result of AI hallucinations, I declined the G15. Ca ^{talk to me!} 11:47, 25 May 2026 (UTC)[]

The citations are the result of this being a "LLM-generated page without human review". It is irrelevant that the same links exist on the Russian Wikipedia, that they don't currently exist because of link rot, or when the Russian Wikipedia text was written. Being a machine translation does not somehow exempt the text from being machine-generated, it is just a subtype of machine-generated text. --gurkubondinn 11:51, 25 May 2026 (UTC)[]

But this doesn't matter anymore (for this particular page) since I've dealt with it by stubifying the article (using a new source). I don't have any strong feelings about if some Russian video blogger has an article on enwiki or not, my only concern is if that page is AI-generated or not. The current revision isn't AI-generated anymore, so the problem has been fixed from my perspective. It just took more time and effort to have to do both things, since compiling these G15 nominations are also quite a bit of work. Both of those probably took more time than the initial creation, which is the part that probably irritates me the most about this whole thing. --gurkubondinn 11:54, 25 May 2026 (UTC)[]

Ah I was misunderstanding your rationale--looks like we have different interpretations of G15. Since I am still new to admin side of CSD work, I'll hold off on declining CSDs from you. In any case, I appreciate for your work here. It's thankless work and people like you help Wikipedia clean from AI slop. Ca ^{talk to me!} 12:46, 25 May 2026 (UTC)[]

No worries! Like I said, the article has been taken care of via stubification instead. It's also slightly more rewarding to "fix" something like this than to get it straight-up deleted, so it's nothing to worry about. I just wanted to get my rationale across since I suspected that there might have been some slight misunderstanding here. Congrats on passing AELECT! --gurkubondinn 12:53, 25 May 2026 (UTC)[]

Been going down the list of the main space articles that this user has created in article space, see: User talk:DddsDD § May 2026.

@Admins willing to patrol AINB: how do you feel about a partial block from article space (before the list of articles to go through gets longer)? The user stopped editing/creating new articles shortly after I got started, but I would prefer if they didn't start again and appended more pages to the list of articles to get through. There has been no communication from the user.

--gurkubondinn 23:03, 23 May 2026 (UTC)[]

While they've stopped for now, communication is required and they have been warned about article-space issues repeatedly before. I've blocked from article space, block can be lifted if communication resumes + reassurance that AI won't be misused to generate content in the future. Chaotic Enby (in solidarity · talk · contribs) 23:18, 23 May 2026 (UTC)[]

Good job with the speedy deletion rationales, gurkubondinn. I deleted the ones you nominated. There are still a ton of article creations from this user to sift through. ~Anachronist (who / me) (talk) 01:10, 24 May 2026 (UTC)[]

Thank you, I'll continue tomorrow with them. I saw you had deleted most of them and ended up reading your talk page...

I wish I didn't have to do anything with AI cleanup, but unfortunately it has become an unpleasant necessity. I often feel like we're engaging in a futile effort, like trying to prevent a rising ocean tide.
— User:Anachronist 11:29, 7 April 2026 (UTC)

Couldn't have said it better myself. I should probably frame this or something.

--gurkubondinn 01:19, 24 May 2026 (UTC)[]

Took care of a couple, looking at some of the remaining articles I'm not seeing as many prose issues, but the sources themselves appear to be the strongest indicator of LLM use. GPTZero flagged most of Yung Trappa as human, for example (if that's worth anything, I'm still getting the hang of manual review per AISIGNS), lending me to believe that much of the user's work was built using LLM assistance, and the sources were likely not checked, but possibly not entirely copy-pasted. ASUKITE 15:00, 24 May 2026 (UTC)[]

GPTZero isn't great at identifying AI prose when sources are cited. I've been skeptical that it checks for hallucinations. Does it? I do know that if you paste in the article prose, it should be formatted as plaintext without any of the citation superscripts (like ^[2] and so forth). If you clean up the wiki-stuff from the text, then ChatGPT can give you a different result. I had one example that it deemed to be likely human flip to likely AI when I stripped out the citations and inserted blank lines around headings. I haven't tried pasting in raw wiki-source, however, because I thought that would confuse GPTZero even more. ~Anachronist (who / me) (talk) 15:46, 24 May 2026 (UTC)[]

GPTZero has a paid feature to check for hallucinations, but I've never used it and I don't know anything about it. --gurkubondinn 16:17, 24 May 2026 (UTC)[]

I’m hesitant about it. I’ve found that it tends to correctly identify prose that has obvious AI signs matches based on our own guides, so I’ll sometimes use it as a starting point, but it will sometimes miss content that is more subtle. ASUKITE 17:58, 24 May 2026 (UTC)[]

As far as I can tell, it appears to miss the obvious AI-generated vapidity of making assertions about the existence of coverage without actually describing the coverage. That's easy enough for humans to detect, however. ~Anachronist (who / me) (talk) 18:21, 24 May 2026 (UTC)[]

• I don't use tools such as GPTZero, for several reasons, including the fact that in my experience any use of AI that it can manage to identify is almost always so blindingly obvious that GPTZero isn't necessary. I also find it very strange that so many people who wish to check for AI use because AI can't be trusted trust an AI tool to do the job for them. This is not just some amusing little paradox: it is a serious concern. I also find it a matter of very serious concern to see that an administrator who chooses to put work into considering cases on the AI noticeboard needs to have it pointed out that an article created by an AI translation tool is an article created by an AI tool, and it makes no sense to say that G15 doesn't apply because the AI-caused problems are just due to unreviewed use of an AI translation tool, as though somehow an AI translation tool doesn't count as an AI tool. JBW (talk) 00:08, 27 May 2026 (UTC)[]

  I'm only just barely starting to learn how to identify AI content myself, having mostly avoided it. I will say the above points have already turned me away from GPTZero and other such tools, if not for their questionable value and reliability, but also nagging interface (as with any modern web service, they have to be in your face constantly asking for money). AI detection tools for generated images do seem to have merit in some cases, but with text it is a waste of time.

  I don't always find it obvious when something is written by an LLM myself, and machine translation tools have already been a problem for a while. Traditional ML-based translations are usually very easy to spot, thankfully. I will just keep reviewing AISIGNS and testing my own judgement. ASUKITE 00:24, 27 May 2026 (UTC)[]

  That's basically what I do, use my own judgment. I find GPTZero to be useful (it provides an additional data point) when I am not certain of my judgment, because some of the minor "tells" may be present but not the primary content-based ones (being unsourced sentences and interpretations or synthesis of primary sources). ~Anachronist (who / me) (talk) 04:49, 27 May 2026 (UTC)[]

Only one article remains, Anna Peresild. It is hard to tackle because there have been substantial later edits, and most sources are not in English. It has been marked as AI-generated with a detailed message. Is that all we can do at this time? Its entry mentions the possibility of AfD but I can't do a reasonable WP:BEFORE on something with mainly non-English sourcing. M kuhner (talk) 03:18, 16 June 2026 (UTC)[]

Yep, that's precisely why I haven't gotten further with that one. It's a tricky one for sure, and I would obviously like to be able to finish cleaning up after this user. But gotta be realistic too, and barring a Russian-speaking editor showing up and helping out, I don't think there's anything more that we can realistically do. ‑‑gurkubondinn 09:17, 16 June 2026 (UTC)[]

While it's a pain, I run the source through Google translate; having an extension in my browser that does this automatically (some of the time) helps. I also have User:Headbomb/unreliable.js in my common.js page to identify potentially unreliable sources, and I check those first. It highlighted references 10 and 22 as possibly unreliable. ~Anachronist (who / me) (talk) 13:48, 16 June 2026 (UTC)[]

I previously saw issues with this editor's work and warned them on their user talk page. I think recent contributions (e.g., [36]) to Balen Shah show signs of AI. Wracking ^talk! 16:50, 25 May 2026 (UTC)[]

The user has been renamed to Renamed user 7779156af9e40e8b3eb3d500bbd466c6 (talk · contribs) and locked. Seems to be their own request. - Ivan530 (Talk) 02:02, 1 June 2026 (UTC)[]

Nevermind, take a look at § Shree Tin, seems to be their new account. - Ivan530 (Talk) 02:19, 1 June 2026 (UTC)[]

Has the cleanup at § Shree Tin dealt with this user as well? Can this one be closed as redundant? M kuhner (talk) 04:43, 13 June 2026 (UTC)[]

Alas. His first single, "Sadak Balak" (Street Kid), was recorded in 2012 from lyrics he had written while in school. The track addressed urban social issues in Kathmandu and gained attention within Nepal's underground hip-hop scene. in the article Balen Shah, which was added in the diff given by Wracking, is not supported by its reference. If I am reading the diffs right, this text has been removed and re-introduced. I can't really follow what happened. But it's there now (I just tagged as failed verification) and I suspect other LLM text remains. Can someone with better diff skills help here? M kuhner (talk) 05:06, 13 June 2026 (UTC)[]

Looking for advice here. Editor13002 (talk · contribs) has made contributions which read as LLM-generated to me, and which get high scores as such if I feed them through zerogpt. Another editor has warned them about this, and I followed it up after they responded to me with text that I thought part-AI-generated. The editor has replied both to the other editor and to me to say that AI was not used. Since then, they have created an article, Matthew Z. Robb, which again reads LLM to me, and is scored 87% AI by zerogpt. I'm unsure of the line between AGF and trying to identify LLM content. I realise machine identification of LLM content can be incorrect. Like another editor above, I'm wondering, is it reasonable to ignore the zerogpt analysis and AGF that this is not LLM content, or should I continue discussing this with the editor? Or a third option I haven't thought of. Tacyarg (talk) 08:31, 27 May 2026 (UTC)[]

Matthew Z. Robb was absolutely created with the help of an LLM tool, and repeatedly and indignantly lying about it is not a good sign, especially since there also seem to be COI concerns. JTtheOG (talk) 08:35, 27 May 2026 (UTC)[]

Rudolph Charles "R.C." Rieder as well. Janet Elizabeth Rieder-Henderson is straight-up 100% AI slop. It's sad to see people's life story reduced to a lazy ChatGPT prompt, but alas. JTtheOG (talk) 08:41, 27 May 2026 (UTC)[]

Dean A. Robb also, father of the previous. Einsof (talk) 14:53, 27 May 2026 (UTC)[]

Also, AI detection tools are not reliable indicators of LLM content and I strongly caution against using one to create a presumption of LLM use. Some possible signs that stand out to me are (1) the appearance of "Legacy" sections in some of these articles, and also (2) the user's very first edit is remarkably sophisticated for a new editor, although it doesn't preclude the possibility of having edited previously without an account or under a different account, or having copied over the source of an existing article as a template. Einsof (talk) 15:08, 27 May 2026 (UTC)[]

Thank you for keeping the discussion grounded and focused on verifiable information. Editor13002 (talk) 20:25, 27 May 2026 (UTC)[]

I commend your tentativeness, but I think it’s best that we stay grounded and avoid becoming emotional. It’s not fair to newer users when those with platform seniority can undermine credibility simply because they disagree with or dislike a response. Editor13002 (talk) 18:10, 27 May 2026 (UTC)[]

One strategy (though it's painfully time-consuming) is to check sources. There are two typical LLM source errors: broken sourcing (dead link, link to wrong source, different parts of cite don't match) and citing something that exists but doesn't say what the text claims it does. If someone wants to defend either of those as not LLM, they are just paying out rope for a finding of WP:CIR because that says they can't source things with basic competence. (When evaluating dead links, take citation age into account. A citation with date accessed in May 2026 and a dead link is likely to be LLM, or a typo. More than a few of these, it's LLM.) M kuhner (talk) 17:03, 27 May 2026 (UTC)[]

I wonder if Editor13002 is using Word's copilot spellcheck or Grammarly without realizing it's AI- this seems to keep happening to others as of late. Sarsenet•he/they•(talk) 18:08, 27 May 2026 (UTC)[]

I can confirm that Grammarly is the tool I use for proofing. I also think it’s important that others participating in this discussion recognize that Wikipedia does not maintain a formal prohibition on the use of AI-assisted tools. What ultimately matters is whether the content complies with Wikipedia’s core policies regarding verifiability, neutrality, sourcing, and accuracy.

I strongly disagree with the idea that individuals with greater platform seniority or administrative influence should be in a position to speculate or imply a false narrative about another editor’s intentions or credibility without evidence. I appreciate your willingness to remain open-minded and grounded in policy rather than assumption. Editor13002 (talk) 18:28, 27 May 2026 (UTC)[]

I also think it’s important that others participating in this discussion recognize that Wikipedia does not maintain a formal prohibition on the use of AI-assisted tools.

Oh, but we do: WP:NOLLM. It even specifically mentions the use of Grammarly.

Now I am going to assume that you simply weren't aware of this, but why did you say it with such utter confidence? --gurkubondinn 20:36, 27 May 2026 (UTC)[]

Unfortunately I'm going need to withdraw from the discussion. This has become emotionally charged rather than constructive. I wish you a wonderful day. Editor13002 (talk) 20:58, 27 May 2026 (UTC)[]

If those AI-assisted tools veer into the territory of generating text, that is absolutely prohibited on Wikipedia, which makes this case all the more troubling. Are you really asserting that you yourself wrote passages like:

• Rieder-Henderson's life combined geology, education, aviation, and scientific research during a transformative period in American earth sciences and aerospace development. Her work reflected the expanding participation of women in technical and scientific professions during the mid-20th century.
• A 1962 scientific directory published in the journal Micropaleontology listed "Janet Rieder Henderson" among professional workers engaged in ostracode and micropaleontological research
• Court filings in federal litigation have identified Robb as class counsel in environmental and consumer-related class-action cases.

It should also be noted that many of these pages were posted mere minutes apart, another classic sign of rapid LLM generation. JTtheOG (talk) 20:39, 27 May 2026 (UTC)[]

They are pre-written in a word processor, reviewed, and then published manually. More importantly, your concerns continue to rely on speculation rather than identifying any specific factual inaccuracies or policy violations. The discussion appears increasingly emotionally charged rather than evidence-based.

Wikipedia’s guidance regarding AI focuses primarily on preventing hallucinated or fabricated information. It does not prohibit the use of AI-assisted tools altogether, particularly when the content itself remains verifiable, properly sourced, and subject to human review and editing. Editor13002 (talk) 20:55, 27 May 2026 (UTC)[]

I reviewed WP:NOLLM on the very first occasion. There appears to be some difficulty on your side interpreting it accurately. You continue to suggest restrictions or conclusions that are not actually stated in the guidance. I've read the entire framework and more than anything you could be hallucinating. Editor13002 (talk) 21:13, 27 May 2026 (UTC)[]

[T]he use of LLMs to generate or rewrite article content is prohibited. How much clearer can that be? JTtheOG (talk) 21:20, 27 May 2026 (UTC)[]

Interpreting Wikipedia policy is like placing what is and is not allowed through a filter. You still have to read the entire policy and apply it in context to develop a clear understanding. Otherwise, you risk drawing conclusions about what is prohibited without fully understanding what is actually permitted. Case in point. Editor13002 (talk) 21:23, 27 May 2026 (UTC)[]

I'm not drawing conclusions. I'm well aware of the guidelines. LLM-assisted generation of text is prohibited. Once again, are you asserting that you yourself wrote the three sentences above? JTtheOG (talk) 21:25, 27 May 2026 (UTC)[]

I'd also point out this talk page message, which reeks of LLM as well. Just another coincidence I suppose. JTtheOG (talk) 21:27, 27 May 2026 (UTC)[]

“Reeks of LLM” is pure speculation. You may believe that statement strengthens your argument, but in practice it weakens it. It reinforces my concern that the discussion is becoming increasingly speculative rather than evidence-based.

It is also important to recognize that people interpret and communicate through the lens of their own experiences and backgrounds. My writing style reflects a legal and administrative background, which naturally results in formal, structured, neutral, and highly procedural language. Bureaucratic or institutional tone should not be conflated with evidence of AI authorship. Editor13002 (talk) 21:33, 27 May 2026 (UTC)[]

"Providing additional evidence of wide use of LLM-like language weakens my argument." For someone claiming to be a lawyer, that seems like a pretty poor argument. Alas, anyone can go read it themselves. This response itself looks AI-generated to me as well. I'll leave this to an admin. JTtheOG (talk) 21:44, 27 May 2026 (UTC)[]

You have not provided conclusive supporting evidence. You have speculated repeatedly, and you continue to rely on speculation rather than demonstrating identifiable policy violations or factual inaccuracies.

There is also nothing in my response where I explicitly stated that I am a lawyer. You inferred that on your own, which further illustrates my concern regarding assumptions being treated as conclusions.

Quite frankly, unsupported speculation and subjective interpretation can be more problematic than AI itself, because at least AI concerns can be evaluated through evidence, sourcing, and verification standards. Editor13002 (talk) 21:53, 27 May 2026 (UTC)[]

Actually they can't, at least not without inventing a time machine (and teleporter) and watching you make the edit, because it is not only about sourcing and verification (though that is something that LLMs do routinely fail at). Our NOLLM guideline is not focused on hallucinations or verifiability, it prohibits the inclusion of LLM-generated content in general, regardless of sourcing and verifiability. To a certain extent, we put a lot of faith into (mostly) completely anonymous users being truthful and honest with us.

Wikipedia has been around for 25 years by now, and it was all researched and typed out by people. The community has largely decided that this is an attribute that we value and want to keep, and has more or less rejected the use of AI in the encyclopaedia. This is something that matters a lot to many of us.

FWIW, when you say that you have "a legal background", that makes me think that you are saying that you are a lawyer. But this is a somewhat vague statement and I am not a lawyer, and I have never worked in law, so there may be some nomenclature that I am not used to. Probably best to avoid making vague statements though, especially if you want to refer back to them later. But I don't really care about what your background is, and this should not be understood as me asking for more details on it. I just wanted to address this part, because it felt a bit off to me. --gurkubondinn 22:23, 27 May 2026 (UTC)[]

• Wide use of LLM-like language (even if you didn't use AI, it's concerning how similar the tone is, esp. when the content is unsourced) in articlespace, talk page, and noticeboards. Detection tools give a 99% AI score, for example, on the aforementioned talk message (and others). But most importantly it's suspicious to the human eye, especially the three sentences I listed above.
• rapid creation of articles containing such text within minutes of eachother
• your first four edits being four brand new articles created from scratch
• deflections from the direct questions asked, such as wether you wrote those three sentences yourself
• Responses in-line with AI use (WP:NOPROOFOFAI, WP:AIWHERE?, WP:THERESNOAIPOLICY (again, very concerning how confidently you inaccurately described the guidelines)
• The appearance of WP:COI/WP:UPE issues

JTtheOG (talk) 22:37, 27 May 2026 (UTC)[]

When did you read it, and have you checked back more recently? Previously it was known as WP:NEWLLM and only prohibited using an LLM to create new articles from scratch. In March of this year, the guideline was revised by the community to prohibit the use of LLMs to generate article contents in general, and became known as NOLLM. And now Grammarly is specifically mentioned, in an explanatory footnote listing a few common LLMs. @JTtheOG is absolutely correct in their interpretation of the guideline, but you may be thinking of an older version of the same guideline. --gurkubondinn 21:32, 27 May 2026 (UTC)[]

Forgot to mention; as far as I can tell, they also are not drawing any conclusions about the guideline. --gurkubondinn 21:35, 27 May 2026 (UTC)[]

Forgive me, I didn't bring my highlighter. You'll have to do the work.

"ext generated by large language models (LLMs) often violates several of Wikipedia's core content policies. For this reason, the use of LLMs to generate or rewrite article content is prohibited, except for basic copyedits and translation of material from other language Wikipedias as outlined below.

1. Editors are permitted to use LLMs to suggest basic copyedits to their own writing, and to incorporate some of them after human review, provided the LLM does not introduce content of its own. Caution is required, because LLMs can go beyond what is asked of them and can change the meaning of the text such that it is not supported by the sources cited. Examples of basic copyedits include spelling, punctuation, and capitalization.
2. Editors are permitted to use LLMs to translate articles from another language's Wikipedia into the English Wikipedia, but must follow the guidance laid out at Wikipedia:LLM-assisted translation.

The writing style of some editors may appear similar to the output of LLMs. The imposition of sanctions requires evidence beyond basic stylistic or linguistic indications. When evaluating possible LLM use, it is best to consider the full pattern of the editor's recent edits and whether the edits comply with core content policies. Editor13002 (talk) 21:38, 27 May 2026 (UTC)[]

So I didn't want to mention this because I thought that it was going to be irrelevant, but I was actually one of the editors involved in drafting the March revision of the NOLLM guideline, so I have a decent handle on what it ended up on saying.

The operative part is the bolded the use of LLMs to generate or rewrite article content is prohibited, and a careful reading of the except will reveal a few things. You are permitted (but discouraged) to use LLMs to suggest[emphasis added] provided that it does not introduce content. The thinking here is that you would ask/input a query to the LLM to suggest changes, and then go back and make those changes yourself (as opposed to copypasting them over). The examples of what counts as basic copyedits in this context are given as spelling, punctuation, and capitalization, which is intentionally rather limited.

The part about how someone's style of writing may appear similar to the output of LLMs is about applying sanctions, which has never been done purely based on stylistic indicators. Quite often, editors that are using LLMs end up being blocked for reasons other than specifically their use of LLMs. Someone is only blocked when it has been deemed to be an action that is necessary in order to protect the encyclopaedia, people are (somewhat famously) not blocked as a punishment. The goal is always to try to work with users and to help them to contribute constructively to the project. When there are concerns weather some editor may be using an LLM for their edits, their supposed LLM use is not always the only concern. Quite often, there are additional concerns around how they respond to being asked about it and how they communicate following being asked about it. For example, sometimes editors become a drain on the most important resource of the community, the free time that people choose to volunteer to the project. Remember that the primary concern is always about if someone is contributing positively towards the project.

Now I would also like to explicitly point out that I have not offered any opinions on weather or not I believe that you are using AI assistance in your edits. I have not taken a close enough look myself to hold an opinion that I want to share. I have made no accusations, I have only offered explanations on guidelines and interpretations thereof.

PS: You can use the {{highlight}} template when you forgot your highlighter at home. --gurkubondinn 22:05, 27 May 2026 (UTC)[]

I'm glad you helped. I understand what the guideline says. My point is that much of this discussion has moved beyond policy interpretation and into subjective speculation about tone, writing style, and assumptions regarding intent.

If there are factual inaccuracies, fabricated citations, or identifiable policy violations, point to them directly. Otherwise, repeatedly asserting that something “sounds like AI” is still speculation regardless of who drafted the guideline.

The issue is not whether LLM-generated article content is prohibited. The issue is whether conclusions are being drawn based on evidence or based on personal impressions and emotional interpretation of writing style. I also never explicitly stated where I forgot my highlighter. I simply stated that I forgot it. So, there's that.

Editor13002 (talk) 22:19, 27 May 2026 (UTC)[]

I am afraid that isn't how it works. A lot of our LLM detection is in fact based on tone and writing styles. In fact, we have a (rather long) document that compiles some of this. We are not only concerned about factual inaccuracies and fabricated citations (though we very much are also very concerned about those). We are concerned about the use of LLMs in general, and also LLM use that does not introduce those specific problems. There are many reasons for this, for instance because of the asymmetric workload that it creates for volunteers. As you have correctly identified, LLM-generated text is prone to generating factually incorrect text (or "filling in the blanks", or "making shit up") but it can also be done at a volume that volunteers editors, who's donated time is our most precious resource, cannot keep up with. Just because someone has made a few LLM-generated edits that were free of factual errors, there is absolutely no guarantee that the next one will also be. Someone would have to check those subsequent edits, which takes up an enormous amount of time because of how fast it can be generated. These algorithms are also prone to generating text in a tone (yes, it is also about tone) that is unsuitable for an encyclopaedia (which I personally think is a product of how much marketing and spam is on the internet). The community has largely decided, by consensus, that we don't want this kind of text regardless of the presence of factual inaccuracies or not.

The issue is that LLM-generated article content is prohibited. You have stated that you did use Grammarly (and word processors, most of which include AI features baked in these days), and we are patiently pointing out to you that this is not allowed on English Wikipedia. Personally, I have even said that I was going to assume that you simply were unaware of this at the time that you made those edits. --gurkubondinn 22:39, 27 May 2026 (UTC)[]

Yes, and that is exactly my point. You closed by stating that you were “going to assume” I was unaware of the policy at the time I made the edits, which still carries an underlying presumption that prohibited AI use occurred in the first place.

There was never a direct admission of AI-generated article writing on my part. I stated that I draft articles in a word processor prior to publishing them and that Grammarly assists with proofreading. Those are not the same thing as explicitly stating that article content was generated by an LLM.

I reviewed the policy, acknowledged the concerns surrounding hallucinations and generated content, and repeatedly pointed back toward verifiability, sourcing, neutrality, and factual accuracy. At no point did I explicitly state that I was using AI to generate encyclopedia content.

That is why I continue drawing a distinction between evidence-based evaluation and conclusions built from inference, tone analysis, workflow assumptions, and interpretive speculation. Editor13002 (talk) 22:48, 27 May 2026 (UTC)[]

No, it does not. It means that I am assuming good faith on your behalf, which is our most important behavioural guideline. I have made no comments or assumptions about your edits, I have only been explaining guidelines and responding to your conduct in this thread. Regardless of weather or not you used (or did not use) LLMs in your previous edits, my assumption is that you were not aware of NOLLM prior to this conversation. Now I will assume that you are aware (as evidenced by you pasting it into the thread), and that none of your future edits will contain any LLM-generated (or LLM-modified) text, regardless of such text having been present (or not present) in your previous edits.

In Diff/1356430518, you said that you could confirm that Grammarly is the tool [you use] for proofing. This is what I meant that you have stated that you did use Grammarly. My recommendation to you is to simply give us some assurances that you will not use Grammarly for your contributions going forward, as that would explain most of the concerns that have been brought up by others in this discussion. Because you are still missing the point when you talk about the concerns surrounding hallucinations and generated content and verifiability, sourcing, neutrality, and factual accuracy. Yes those are important (very important in fact), but so is not using LLMs. --gurkubondinn 23:06, 27 May 2026 (UTC)[]

You also continue to avoid acknowledging that Grammarly includes non-generative proofreading modes. Basic grammar correction, punctuation review, spelling assistance, and syntax cleanup are not inherently the same thing as LLM-generated article writing.

That distinction matters because the policy is aimed at generative content creation and rewriting, not the mere existence of proofreading software or conventional editorial workflows. Treating every mention of Grammarly or a word processor as presumptive evidence of prohibited AI use oversimplifies the policy and ignores the distinction between traditional editing assistance and substantive content generation. Editor13002 (talk) 23:10, 27 May 2026 (UTC)[]

The use of Grammarly is prohibited on the English Wikipedia by the NOLLM guideline. I believe that Grammarly are not super forthcoming about the role of AI in their product, but you can read some about it on the Grammarly article in the encyclopaedia. I'm not here to quibble about various modes of this software. --gurkubondinn 23:15, 27 May 2026 (UTC)[]

I'm not here to quibble either. To be clear, I have not located language within the guideline that explicitly states “Grammarly itself is prohibited.” The policy discusses the use of LLMs to generate or rewrite article content and references Grammarly because some versions and features now include generative AI functionality.

However, Grammarly also contains non-generative proofreading and grammar-correction functionality. Those features are materially different from using an LLM to generate encyclopedia prose or rewrite substantive article content.

That distinction matters. Otherwise, the interpretation becomes so broad that ordinary proofreading workflows, traditional grammar correction, spelling review, punctuation cleanup, and even standard word processors become subject to presumption simply because modern software ecosystems now contain optional AI integrations.

My point has consistently been that the existence of proofreading software alone is not direct evidence that prohibited generative AI content creation occurred. Editor13002 (talk) 23:19, 27 May 2026 (UTC)[]

That specific language is in the first footnote in WP:NOLLM § Notes (#cite_note-1). If there is a non-generative mode of Grammarly, and you are only using this mode, then that does not explain the concerns raised by others in this thread. As such, people are going to want to have that adequately explained as you are saying that this is not down to your use of Grammarly (which we have seen cause similar issues with other editors). And yes, this is a problem with "modern software ecosystems", but that is an issue that you have to take up with the vendors of said software. Unfortunately, software is getting worse, much to my chagrin. I recommend FOSS software, which usually gives you more control and more choice. Personally, I do most of my typing in Vim, neovim and Firefox (which has an excellent spell checker). --gurkubondinn 23:30, 27 May 2026 (UTC)[]

Both aspell and Hunspell still work just as well as ever. They are both excellent pieces of software, and both are actively maintained. --gurkubondinn 23:33, 27 May 2026 (UTC)[]

I’m more than willing to switch to one of the applications you suggested if it resolves the concern and closes the discussion entirely. That includes removing or refraining from adding AI-related warning templates to my articles absent identifiable policy violations or demonstrable evidence of prohibited conduct.

If editors dislike my writing style or tone, they are welcome to contribute constructively to improve the articles themselves. That is how collaborative editing is supposed to function. What I am not open to is being treated with suspicion or criticized simply because my writing style does not resemble yours.

Until consensus is reached through proper discussion and policy-based reasoning, I will continue removing warnings and templates that I believe are unsupported, speculative, or improperly applied. That includes AI-generation notices as well as criticisms such as “written like a resume,” which, if anything, further supports my argument that much of this discussion has shifted toward subjective interpretation of writing style rather than identifiable policy violations or factual inaccuracies. Editor13002 (talk) 23:40, 27 May 2026 (UTC)[]

What you seem to be asking is for me to abandon proofreading software altogether simply because some versions of it contain optional AI-related features. For your assurance. That is not a reasonable or clearly distinguishable standard. Editor13002 (talk) 23:14, 27 May 2026 (UTC)[]

The policy is specifically addressing the use of generative AI and LLM-assisted content generation or rewriting. It cannot reasonably be applied as a blanket assumption against all proofreading software or all word processors universally, particularly in situations where non-generative proofreading modes are being used or where the software itself contains no AI-generation functionality at all.

Otherwise, the standard becomes unworkably broad. Basic grammar correction, spelling assistance, punctuation review, and conventional editing tools have existed long before modern LLM systems. Conflating traditional proofreading with prohibited generative content creation removes the distinction the policy itself attempts to make between basic copyediting assistance and substantive AI-generated writing.

That distinction matters because the presence of a word processor or proofreading workflow alone is not direct evidence that article content was generated by an LLM.

Maybe when I’m not being continuously bombarded with accusations and suspicion, I’ll consider assisting with that. Editor13002 (talk) 23:00, 27 May 2026 (UTC)[]

Grammarly is not "traditional proofreading". Therein lies the the problem. --gurkubondinn 23:09, 27 May 2026 (UTC)[]

Well, I forgot my highlighter at home. Where I am right now, but I can't find it, so it isn't of much use anyway. I'm not even sure that I have one. --gurkubondinn 22:42, 27 May 2026 (UTC)[]

Adding yet another egregious example to the list: A 1971 industry publication described Continental-Emsco as a manufacturing and marketing supplier serving the oil and gas industries and referenced Rieder among the company’s executive leadership JTtheOG (talk) 23:50, 27 May 2026 (UTC)[]

That's not a valid argument. I removed the notice. Editor13002 (talk) 23:52, 27 May 2026 (UTC)[]

Perhaps everyone here can agree that stylistic features alone do not prove the use of LLMs. At the same time, this is not a deposition, and candid communication would be appreciated and may help to dispel whatever atmosphere of suspicion may hang here. Most editors are not interested in keeping track of whether the precise wording of a particular talk page comment did or did not function as an admission of something, or whether an edit can or cannot be proved to some evidentiary standard to have violated the letter of some policy. To the extent that it takes up more community resources to parse through, that kind of communication is actually a disadvantage here. Einsof (talk) 23:53, 27 May 2026 (UTC)[]

I agree, which is why I provided more than that. JTtheOG (talk) 23:54, 27 May 2026 (UTC)[]

Before I log off, also very interesting how their edit summaries when removing AI tags read No fabricated content, hallucinations, or false citations identified, once again missing the point entirely that even if these were not present, LLM use is not allowed to generate content. Point, blank, period. JTtheOG (talk) 00:20, 28 May 2026 (UTC)[]

“Identified” is a telling choice of word. If you write and cite something yourself, you know for a fact whether you included any “false citations” because you’re the one who cited them; you can’t accidentally include them unless you didn’t actually read them. You don’t need to do an additional “identification” step on your own work and your own memory. Gnomingstuff (talk) 08:58, 28 May 2026 (UTC)[]

@Admins willing to patrol AINB: I believe that these examples are LLM-generated, contrary to this user's repeated claims. There are also other concerns

• COI: they have claimed to represent an organization [37] but then dissembled about the details [38][39]
• Repeated large content removals [40][41]
• Prolific wikilawyering over WP:NOLLM
• Repeatedly removing {{ai-generated}} tags from articles they created, apparently both while logged in [42][43][44][45] and out [46]. Note the edit summary used by the TA here [47], which is identical to some of Editor13002's.

I think there is enough going on to merit admin attention. NicheSports (talk) 00:06, 28 May 2026 (UTC)[]

Looks like they were temporarily blocked for edit warring just as I posted this NicheSports (talk) 00:08, 28 May 2026 (UTC)[]

Thanks, NicheSports. I was just about to comment re possible CoI issues (eg uploaded photographs of article subjects "from Family album" / "from Family Photo Archive") and use of poor-quality sources (a blog, which also reads part LLM-generated, at denncoholdingcompany is cited in all five of the surviving articles created by Editor13002). I asked the question that led to this discussion; thanks to the other editors who have answered here, it has been helpful in considering how to approach possible LLM-editing in the future. Tacyarg (talk) 00:24, 28 May 2026 (UTC)[]

One thing I've noticed (pardon me if it's off-topic) is that AI tends to interpret suspicion of AI-use and being challenged as some sort of "emotionally charged" aggression and takes it rather personally - insofar as AI can take things personally. It'll try to "refocus" on other things, or ask you to point out what looks like AI (for future training?).

If you continue to politely but sternly push back on this, it will refuse to engage further as per the AI's programming not to get involved in emotional arguments (even if there isn't one in reality).

The fact that this editor said they'd stop responding due to nonexistent aggression then continued to post, along with other AI signs (including the classic "not X but Y") make me 99% sure that AI was being used during this discussion. Blue-Sonnet In solidarity 14:04, 30 May 2026 (UTC)[]

I have looked at the two surviving articles of this editor (in cleanup list above). Both have source/text mismatches and are also substantially sourced to blogs. However, I think that if LLM issues were put aside, Dean_A._Robb might survive an AfD as there's reasonable evidence for notability. (Bios are not my strength, though.) I don't feel I can PROD them as it is not uncontroversial. I could put them on AfD, or we could build an LLMPROD consensus. Opinions? M kuhner (talk) 05:45, 13 June 2026 (UTC)[]

User:AllyMcDonaldAlonso has repeatedly recreated LLM-generated articles on non-notable related subjects, indicating a probable case of COI or UPE. They are obviosuly WP:NOTHERE and should be banned asap, in my opinion. Thank you, JTtheOG (talk) 08:32, 27 May 2026 (UTC)[]

👁 Image
Moved from § What is reasonable to do when text is identified by AI as LLM-generated, but editor insists it is not?

 – --gurkubondinn 09:05, 27 May 2026 (UTC)

Hi there - i havent used LLM - i have written the document in a word document, which i can share - it is using external links that i have referenced in the guides i read on Wiki... is there anything i can do to update to show it isnt llm AllyMcDonaldAlonso (talk) 08:35, 27 May 2026 (UTC)[]

sorry i replied to the wrong thread AllyMcDonaldAlonso (talk) 08:41, 27 May 2026 (UTC)[]

I've taken the liberty to move your reply from the other thread. --gurkubondinn 09:05, 27 May 2026 (UTC)[]

I have only created two articles, and i havent used LLM generated text - i wrote it using guides from Wiki about how to ref sources AllyMcDonaldAlonso (talk) 08:42, 27 May 2026 (UTC)[]

Are you claiming you personally wrote such gems as Coverage of the company's strategy during this period noted the growth of digital media consumption and international television markets and Industry commentary has described Sandbox as part of the growing market for digital learning and children's media platforms. ? If so, that's extremely concerning as well. JTtheOG (talk) 08:45, 27 May 2026 (UTC)[]

i was trying to make sure it was related to the the article text from the external sources that I was referencing - i thought that was the right was to reference but also make sure it was clear it was independant coverage... i can remove that line and those references completely if you prefer. AllyMcDonaldAlonso (talk) 08:48, 27 May 2026 (UTC)[]

So you are, in fact, claiming to have written this yourself, right? JTtheOG (talk) 08:52, 27 May 2026 (UTC)[]

I checked the references at Bhavneet Singh (currently up for speedy deletion) and at least some appear to be hallucinated. That is, while the links are real, they link to articles that do not bear the same title at all as in the footnotes (which claim Singh is in the headline), or are even from a different publication. So if it isn't LLM work, it's possibly a case of deception in order to inflate the notability of the subject. OsFish (talk) 08:57, 27 May 2026 (UTC)[]

Hi there - which links are wrong - i am just going through the draft page and if i referenced any of them wrongly, that will be my writing of them and i can edit them... I tried to get all the refences right and very happy to edit if i got any wrong as i wrote them. AllyMcDonaldAlonso (talk) 08:59, 27 May 2026 (UTC)[]

i have just found one that is linked wrong - i will make the edit now - sorry it was missing a word in the weblink... AllyMcDonaldAlonso (talk) 09:02, 27 May 2026 (UTC)[]

There's more than one. There are several that are not as claimed in the references. I'm afraid I find it beyond the bounds of credibility that every single one is human error. OsFish (talk) 09:12, 27 May 2026 (UTC)[]

I am not experienced with this at all, so sorry if i didnt get the referrecing totally right - the pages are all real, i just referenced them wrong including having the home page for the world economic forum rather than the actual link to the document - i have however updated that... please do let me know what else i can do to get the submission right - i am learning as i go. AllyMcDonaldAlonso (talk) 09:15, 27 May 2026 (UTC)[]

No, some of the links don't even mention the subject, and the named articles don't exist.

This isn't a technical issue. Either you mislabelled multiple references, or an LLM hallucinated them. OsFish (talk) 09:24, 27 May 2026 (UTC)[]

hi there - which ones done mention him - i have just beent though ever link again and they are all the right ones... is it the PDF where he is mentioned near the very end of the document - he is named on page 13 - but i did worry if that one would be ok? AllyMcDonaldAlonso (talk) 09:27, 27 May 2026 (UTC)[]

The thing is, the history of your edits is visible, as they are for all editors here. So people can see that you’ve gone over and corrected several, although at time of writing, still not perfectly. So it doesn’t look good for you to claim everything was all good apart from one little error.

The reason why LLM use is banned is because it creates work for other users tidying up the hallucinations that LLMs introduce. If someone hadn't noticed the AI writing, and you hadn’t been brought here, and I hadn’t taken my own time going over references, these errors would either have persisted, or some other editor sometime down the line would have had their time taken up tracking down the right links, getting angry or frustrated at whoever it was made this mess in the first place.

So it would have been wiser to be more open. Alas, as often happens, that didn’t happen.OsFish (talk) 09:45, 27 May 2026 (UTC)[]

Im sorry you dont think i have been open - i have been and i am trying to share all my learnings too - i have said i have changed a number of the links as they were wrong - including having the home page of them not the specific one for at least 2 of them. I took the links from the google search, i have tried to make any edits that you have said you dont like - is there anything else i can do to show i am trying to learn and make the edits of the language or links you didnt like... AllyMcDonaldAlonso (talk) 09:52, 27 May 2026 (UTC)[]

@AllyMcDonaldAlonso: please also take this is as the friendly advice that it is meant as. I still haven't taken a close look at your edits. Wikipedia editors are generally a forgiving bunch, who are invested in helping new people get acclimatised to the project and to contribute constructively. We want more people to edit this thing, we want as many people as possible to stick grubby little fingers into articles and plot around as possible (we just want them to use their own fingers, and we want them to make the articles better in some small way). If you did use AI at first, that is not a problem. Seriously. It takes time to learn how this place works, it is different than most of the rest of the internet these days (in a good way, I think), and nobody expects a new editor to know our rules and expectations from the get-go. So if you did use AI first, because you didn't know that it isn't allowed, that is something that will be easily forgiven and moved on from. It would only be a problem if you either 1) are try to be vague/deceptive about it or 2) continue to do so after you learn that it isn't allowed. I still haven't taken a close look at your edits (intentionally at this point, since I want to approach this as neutrally and free-of-preconceptions as possible), so I am not passing any judgements here as to weather you may or may not have used AI. --gurkubondinn 10:00, 27 May 2026 (UTC)[]

hi there - thanks for the comments and i do appreciate the help - i am trying to learn as much as i can. I wrote the document myself in word, i use co-pilot within it for spelling and grammer check as i am dyslexic but i didnt think that was using LLM to create which i did read was not ok- and it didnt change any of my text structure just grammatical errors- if that means i did use co pilot for spelling and grammar was wrong then i do apologise - i thought it meant did i use it to write the whole thing... i think the sentances that have been flagged are my wording that i gook from reading the guidance about independance in the text - i didnt want to say that something was fact, i wanted to say that external coverage said something... i was hoping that was the accurate way of doing it. is there any other advice you can give of what i should do better for the future? AllyMcDonaldAlonso (talk) 10:13, 27 May 2026 (UTC)[]

I was going to ask if you were aware that Word now includes an AI (that can't be turned off as far as I understand). Unfortunately it will always change more than you expect, and you can't rely on it to just fix spelling and grammer. You're better off sticking to plain old spell checkers, which Word has had since at least the early 2000s. I'm pretty sure that Word used to have a "normal" grammar checker many years ago, but it seems to have disappeared (or been a false memory, but I swear that it used to underline some sentences with a jaggedy green line).

Just enable the spell checker in Firefox, and use the Wikipedia editor. Word isn't very conductive for writing Wikipedia texts, since it is a WYSIWYG editor, with it's own formatting tools and formats. You can try using the WP:VISUAL editor if you want a WYSIWYG editor that is specifically made for Wikipedia. If you want to write locally, you can try WP:TEXTES § Notepad++ (assuming you are on Windows since you are using MS Word) for example, but the Visual Editor (or regular Wiki editor) is probably better. If you want to keep a local copy of your work, just copy and paste the wiki source into a text file on your computer. --gurkubondinn 10:36, 27 May 2026 (UTC)[]

this is SO helpful! thank you - i use a mac (office for mac) and didnt think there was a way to turn off copilot for spelling and grammar so i will reserach how to do that - or as you said, just use the wikipedia editor is just easier then you avoid any issues like this. Does this mean this article can pass with the edits we discussed or does it need to move to draft first for me to edit? AllyMcDonaldAlonso (talk) 10:43, 27 May 2026 (UTC)[]

On macOS you could for example use the Atom or Sublime Text editors (or any number of other text editors on the list of text editors right here in the encyclopaedia), and macOS also has a built-in text editor (though I personally don't like it). Can't say much about your specific article, other than if you have a conflict of interest then the page should go through the articles for creation process anyway. --gurkubondinn 10:50, 27 May 2026 (UTC)[]

I would second gurkubondinn that we absolutely want people to edit. But I would also echo the need for openness. Some of the URLs and reference article titles you originally used could not have been cut and paste errors. They resemble either material on various websites or what the article needed them to be, precisely in the manner that an LLM hallucinates such things. LLMS are deeply problematic for Wikipedia content because they are too unreliable and in an unaccountable manner which all in all causes more disruption than it is worth. For example, they often fail to summarise sources accurately, and not in a human error way, but in a “makes things up out of the blue” way. And it’s difficult to get them to stop doing that. There is thus a trust issue if someone is using LLMs to write content.

The article has more issues than LLM use. It needs to establish wp:notability ie that the subject merits an entry. And, given your persistence in trying get articles up about Singh and his company Sandbox, a suspicion of conflict of interest. Which is why openness would be a good idea. OsFish (talk) 11:22, 27 May 2026 (UTC)[]

They "fill in the gaps", weather you ask them to or not. It's what they are designed to do, they are just next-word-prediction algorithms. --gurkubondinn 11:43, 27 May 2026 (UTC)[]

@AllyMcDonaldAlonso: please take this the genuine and friendly advice that it is meant as. I have no opinion on your edits, and I have in fact not even looked at them, so I am not trying to insinuate anything. You should read the WP:USED-AI? essay before you reply further here, especially if you did use AI. Since I have not looked at your edits, I am not saying that you did (or didn't) use AI, just that you should read this essay. --gurkubondinn 09:26, 27 May 2026 (UTC)[]

hi there - thanks for that - really helpful - i appreciate it. I ready the gudiance on how to write a peice so that it was independant - i hadnt seen this before as i didnt know it was a thing - so thank you very much AllyMcDonaldAlonso (talk) 09:29, 27 May 2026 (UTC)[]

Nice catch. I just noticed the machine-generated text and didn't even bother looking at sources. JTtheOG (talk) 09:06, 27 May 2026 (UTC)[]

what do you mean by that - if there is wording you dont like i am happy to ammend it - i have just checked all my links and they all are correct now... i can make any edits you prefer... thank you for your hekp AllyMcDonaldAlonso (talk) 09:13, 27 May 2026 (UTC)[]

I mean that I immediately recognized the above sentences as LLM-generated, and quickly brought the case here without checking the references, as OsFish did. JTtheOG (talk) 09:19, 27 May 2026 (UTC)[]

in one of the artcles referenced it says "He has driven the growth of the Comedy Central" so i used that to say the article noted the growth of digital media" - I know from the guiance you dont want opinion and fact, so i was trying to mirror that. Would you prefer if i removed those two sentances or reworded them - i could say "credited with the growth of digital media brands" AllyMcDonaldAlonso (talk) 09:22, 27 May 2026 (UTC)[]

It would be most preferable if you didn't use AI to publish articles in the first place. Another piece of advice would be not to lie about it. JTtheOG (talk) 09:25, 27 May 2026 (UTC)[]

ok - so i understand with both of those things and totally agree - i was trying to make amends from language you didnt like - i thought noted was a good word which i did write as i thought it sounded indepedent which i know is very important- i am trying to make edits and learn what you dont like... is there anything I can do to help give you the assurance and edits needed - if i remove those two references you didnt like, will that help? AllyMcDonaldAlonso (talk) 09:32, 27 May 2026 (UTC)[]

What do you think of the two sentences that @JTtheOG pointed out above? If you think that they merit being written differently, how would you rewrite them? --gurkubondinn 09:41, 27 May 2026 (UTC)[]

For this one "Coverage of the company's strategy during this period noted the growth of digital media consumption and international television markets" - "Singh was responsible for strategic growth in Central and Eastern Europe, Russia, the Middle East and Africa"

and for the second quote "Industry commentary has described Sandbox as part of the growing market for digital learning and children's media platforms." - i would change to "Sandbox creates digital learning for children under the FutureFit Framework" AllyMcDonaldAlonso (talk) 09:49, 27 May 2026 (UTC)[]

And which versions do you think are better to use in the article, and why? Which one do you think is better for the reader? --gurkubondinn 09:53, 27 May 2026 (UTC)[]

so i would use the new updates as they are more specific and mentioned directly in the articles - i know the text has to be very careful to be totally neutral so if i replaced with those two, would you be happier? AllyMcDonaldAlonso (talk) 09:55, 27 May 2026 (UTC)[]

More friendly and well-meant advice, @AllyMcDonaldAlonso; I think that you should stop asking people if they would you be happier because it comes across as slightly passive-aggressive and not very collaborative. I am assuming that you don't mean it that way, but I feel that you should be aware that this is how it might be received by others. --gurkubondinn 10:03, 27 May 2026 (UTC)[]

oh my gosh - i am so sorry if it comes across as passive aggressive - it is TOTALLY not meant like that at all. i am trying to understand what edits i need to make sure the reviewers would be content with the work - Please take this as someone learning and finding their feet and trying to navigate a complicated process- i was just trying to understand what edits would make the reviewers happier. If i said "does this meet the criteria" - would that be better? Thank you againg for taking the time - i do appreciate it a lot. This is really hard to get right and i am trying my best. AllyMcDonaldAlonso (talk) 10:15, 27 May 2026 (UTC)[]

That's what I thought, so I wanted to let you know. Tone doesn't travel well through writing (except for writers that are much more skilled than I am), but I gathered as much from your other responses here. Something like that would probably be better, or a longer "does rephrasing this address the issue or am I missing a larger problem here?" could also work (but I don't want to speak for you).

And don't worry, this is a pretty complicated place at the end of the day and even the most experienced users won't remember every single policy and guideline, though you eventually start developing a sense for when you might have to look something up and remembering where to look things up.

Keep in mind that writing a new article from scratch is probably one of the most difficult tasks that you can do on Wikipedia. It takes a lot of time and careful work. Writing one sentence isn't just typing up the words in the sentence, it also involves the research to understand what the sentence should say, citing the source (correctly), verifying and researching the source, and then probably reworking that sentence a dozen times. You don't have to start contributing to the project by writing a new article, there are lots of other things that you can do to contribute. You would also learn along the way, and slowly build the skills to write a whole new article. It can take many years for editors to write their first article (I still haven't, but I've been working on one in draft space for a couple of months now). Some never do. And that's fine. --gurkubondinn 10:28, 27 May 2026 (UTC)[]

Recently I gave some similar advice to a different editor in § Could you potentially help me in the future? on my talk page that you might find helpful. --gurkubondinn 10:38, 27 May 2026 (UTC)[]

i didnt even know there was a draft space - thank you so much for telling me - is it possible to move this article to the draft space so i can do that - i thought i had to just create a page for review... that would be so helpful to have an area i can use to test in... AllyMcDonaldAlonso (talk) 10:40, 27 May 2026 (UTC)[]

Yes, that is possible! See for example WP:DRAFTIFY, though that is more geared toward how to move someone else's page into draft space. Just click "Page" and then "Move" (possibly it is under the "Tools" drop down), and you will see the move dialogue: Special:MovePage/Bhavneet Singh. There, just select "Draft" in the drop-down under "New title", to the left of the page title. --gurkubondinn 10:45, 27 May 2026 (UTC)[]

thank you - i have done that - so i assume i can now continue to make edits and then resubmit once they are done? Thank you so so much for taking the time to be kind and helpful - i really appreciate it so so much. AllyMcDonaldAlonso (talk) 10:49, 27 May 2026 (UTC)[]

@JTtheOG: do you want to withdraw the G11 while this is in draft space? --gurkubondinn 11:05, 27 May 2026 (UTC)[]

I still have my reservations, but I suppose so. Thanks for your help here. JTtheOG (talk) 11:24, 27 May 2026 (UTC)[]

Sparks19923 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

See ANI thread for details. Possible AI agent. Cleanup needed. Fermiboson (talk) 23:06, 27 May 2026 (UTC)[]

Even their earliest edits have lots of AI signs. A full cleanup case may be needed. InfernoHues (talk) 01:51, 29 May 2026 (UTC)[]

Now they finally admit that they are using AI. Though somehow it feels like AI response. - Ivan530 (Talk) 02:35, 1 June 2026 (UTC)[]

User was warned in March about AI use, and denied it with the typical "if there are any specific concerns". [49] has AICURLY and emphasized, [50] has highlighted, [51] AIBOLD and edit summary... list goes on. Fermiboson (talk) 13:35, 28 May 2026 (UTC)[]

Will warn regarding this thread. | One Reaction was here. Got a complaint? 20:18, 28 May 2026 (UTC)[]

I was reviewing newly created article Artificial intelligence in Indonesia and feels the tone is off. At first I thought it's because NPOV, but then I notice that it might be AI generated. Especially in the lead Artificial intelligence in Indonesia refers to development, ... and Recent developments include ....

Then I investigate the editor Iryna Odessa UA (talk · contribs), while I haven't found other potential AI edits yet (or maybe failed to notice it), I do find some their edits bizarre. For example in that article, all references are named, even most of them are only cited once. The same happen to most of the article created by them, like in News Diggers!. - Ivan530 (Talk) 06:18, 31 May 2026 (UTC)[]

I haven't dug into their contributions yet but the user page also gives off weird vibes. —In solidarity with Wiki Workers United · ClaudineChionh (she/her · talk · email) 06:28, 31 May 2026 (UTC)[]

I think I found more proof that this user is using AI. On this article, take a look at the fourth references (Book Arsenal). The link lead to 404. Could it be that linkrot happening between 21 January 2026 and right now? no, that's because Wayback Machine archive in 7 April 2026 also lead to 404 (and it's the only archive in that URL). Compared that to the the real link that are archived many times since 2022. So it's very likely that AI hallucinate that link. - Ivan530 (Talk) 16:55, 31 May 2026 (UTC)[]

Yes, it looks like the bulk of their mainspace contributions are composed of lightly edited AI-generated text. Compare this article creation with these two comments [52][53] that were apparently written in their own voice. I also wonder about potential UPE; this is an interesting cluster of articles to have created for a Ukrainian editor: [54][55][56][57][58][59]. I just dropped them a note on their talk page about both. NicheSports (talk) 18:43, 31 May 2026 (UTC)[]

Just to be fair, I name all my references nowadays because it makes my article writing process easier. Not sure this is a particularly strong sign. The 404 on a very recently added link is a much bigger red flag. M kuhner (talk) 23:02, 31 May 2026 (UTC)[]

404 links are a criteria for G15 as they are an implausible or nonsensical reference. Fortek67 (talk) 15:27, 1 June 2026 (UTC)[]

Seems pretty obvious that it's AI, like Frednard Gideon. Good to start tagging the articles with an AI notice. Let's give the user some time to respond to the messages as they haven't edited since they were sent. After that we can consider starting a consensus if we should either mass delete or draft these articles. Fortek67 (talk) 15:26, 1 June 2026 (UTC)[]

There is around 50 articles created by them based on counts in xtools. So it's gonna be pain to tag them. - Ivan530 (Talk) 15:29, 1 June 2026 (UTC)[]

No worries. I'm doing it right now. Can you just quickly send me the link to the user's xtools page? Can't be bothered to find it. Thank you, Fortek67 (talk) 15:31, 1 June 2026 (UTC)[]

here - Ivan530 (Talk) 15:33, 1 June 2026 (UTC)[]

👁 Image
 Done. Near the end I realised I accidently dated it as May 2026, but hopefully that shouldn't be a problem. Thanks for the help btw! Fortek67 (talk) 15:53, 1 June 2026 (UTC)[]

Alright, I just double check and it seems all of their created articles is tagged now. But for University of Indonesia Press the tags should be removed since their article is deleted, then another editor create new one. Unless somebody can confirm by comparing the deleted version and the current one are the same. - Ivan530 (Talk) 16:24, 1 June 2026 (UTC)[]

Good catch. I've removed it, at least for now. As you said it's better for someone to check and compare the deleted version. Fortek67 (talk) 16:28, 1 June 2026 (UTC)[]

Although references 1 and 5 do seem to lead to 404 pages? Though that may just be my browser Fortek67 (talk) 16:30, 1 June 2026 (UTC)[]

1 and 5 are Indonesian government website. They tend to shutdown the server outside office hour (right now it's midnight in Indonesia), or IP block non Indonesian IP. I would check it myself latter and see if it still lead to 404. 1 is for Indonesian Ministry of Education and 5 is the National Defense Library. - Ivan530 (Talk) 16:38, 1 June 2026 (UTC)[]

Good to know! Fortek67 (talk) 16:42, 1 June 2026 (UTC)[]

Nevermind, seems to be AI generated too. Already retagged it. The first ref., SINTA, changed it's URL on August 2025 (instagram announcement) yet it claimed to access it on April 2026. Also the corrected link just lead to list of books published by somebody. So it's very likely AI hallucinated. I don't know if Lit.globe.paperbacks (talk · contribs) is Odessa's sockpuppets or another editor. - Ivan530 (Talk) 17:37, 1 June 2026 (UTC)[]

I was thinking that if it was AI generated, it is probably a socketpuppet account of Iryna. Either that or the bizarre coincidence that it happened to be another LLM account that decided to create the same article.

Now, after doing some further investigation, University of Indonesia Press was originally created on 17 April, then deleted the same day, according to xtools. Lit.globe.paperbacks created their account on 20 April. On the same day, they began making several edits from the suggested edits panel. I assume they started doing this to be granted autoconfirmed perms, because, just over five days later, they created University of Indonesia Press.

So I think the best option now is to open up a socketpuppet investigation regarding this user, and let the administrators take care of it. Fortek67 (talk) 20:14, 1 June 2026 (UTC)[]

If you want to compare the text in the current version of the article to the deleted article, you should be able to request to have a WP:REFUND of the deleted version emailed to you. This should be a valid rationale to get a copy sent to you privately. --gurkubondinn 20:52, 1 June 2026 (UTC)[]

Good to know. Thank you Fortek67 (talk) 15:13, 2 June 2026 (UTC)[]

Turns up both accounts are actually a socketpuppet of Ellie WombatWhisperer – see Wikipedia:Sockpuppet investigations/Ellie WombatWhisperer. All involved accounts have been blocked.

There now seems a lot more cleaning up to do. Looks like Shawn Mangui-Ob never made any contributions, so that leaves us with Iryna Odessa UA, Lit.globe.paperbacks, and Ellie WombatWhisperer. Ellie WombatWhisperer never created any pages but did add LLM content to pages, so I will start removing that.

This leaves us with just Iryna Odessa UA and Lit.globe.paperbacks, whom both mass created LLM-generated articles. If we could get a consensus for a full cleanup per WP:LLMPRV, that would be great. Please comment Support or Oppose below: Fortek67 (talk) 15:22, 2 June 2026 (UTC)[]

Support mass cleanup per WP:LLMPRV. I see no reason to wait. M kuhner (talk) 20:42, 2 June 2026 (UTC)[]

Support applying LLMPRV for full cleanup without waiting. Getting started earlier makes reverting easier and avoid potential waste of time in an already arduous task. --gurkubondinn 20:47, 2 June 2026 (UTC)[]

Support Yep, no reason to dilly-dally when cleaning up confirmed socking. Stepwise Continuous Dysfunction (talk) 21:11, 2 June 2026 (UTC)[]

Alright, I've now went ahead of nominating all articles for deletion created by Iryna Odessa UA and Lit.globe.paperbacks, except Raven's Way, as that seems to have been fixed up by another editor. I also reverted all LLM edits by the two users.

We should be good now, just have to wait for the articles to be deleted. Thank you to @Ivan530 for your report, and to everyone who helped with this case. Fortek67 (talk) 14:50, 3 June 2026 (UTC)[]

Newcomer Tasks strikes again! Gnomingstuff (talk) 15:03, 4 June 2026 (UTC)[]

While I agree with you about Newcomer Tasks in general, this is a sock--not a newcomer. M kuhner (talk) 15:23, 4 June 2026 (UTC)[]

Yeah, pretty unfortune how helpful tools for new editors are being misused. Fortek67 (talk) 15:29, 4 June 2026 (UTC)[]

See Wikipedia:Sockpuppet investigations/Iryna Odessa UA. Fortek67 (talk) 20:22, 1 June 2026 (UTC)[]

Well spotted - your filing turned up a few more. Yet another LLM user operating multiple accounts NicheSports (talk) 03:12, 2 June 2026 (UTC)[]

It has come to Pppery's attention that ~2026-33518-93 is trying to remove the LLMPROD nominations. If someone here can keep an eye on all articles, that would be great. It might be possible that that the temporary account is Lit.globe.paperbacks, so a block may have to be enforced. Fortek67 (talk) 17:01, 7 June 2026 (UTC)[]

I am not that user. Feel free to check if you have more time to waste. And my edits were most constructive. I DID check the sources and content of those pages. ALL of the ones I had deproded. The rollback by Pppperry on the other hand was clearly unjustified. But I have no time to undo each of the articles unduly PRODed (by you). Shame. Anyway, keep up the good work! ~2026-33518-93 (talk) 17:32, 7 June 2026 (UTC)[]

Korii n98 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

They've been messaged three times about AI use but are unresponsive to concerns. ~WikiOriginal-9~ (talk) 12:20, 1 June 2026 (UTC)[]

Consider to start removing and reverting content they add. With them not listening to you, you will probably have to wait for an administrator to respond to this. Fortek67 (talk) 15:36, 1 June 2026 (UTC)[]

Something like a pblock from article space can help WP:TCHY editors locate their talk pages, and this seems to be a user editing from a phone. --gurkubondinn 17:23, 1 June 2026 (UTC)[]

The editor has now been blocked by @Joy. There are a number of large updates they have made recently, I reverted Acid Jazz, the first one I came across by accident. I'm concerned about the Srebrenica article due to the sensitivities on that one. ChrysGalley (talk) 11:35, 2 June 2026 (UTC)[]

Should we gather consensus to apply LLMPRV right away? --gurkubondinn 11:43, 2 June 2026 (UTC)[]

• Support application of WP:LLMPRV - I could find no merit in the AFC drafts or the Acid Jazz edits. ChrysGalley (talk) 11:46, 2 June 2026 (UTC)[]
• Support as having suggested it, and for the same reasons I gave at § User:Daniel G Rego the other day. --gurkubondinn 11:49, 2 June 2026 (UTC)[]
• Support. This type of situation is one where we are in a hurry, because it gets so much worse if other edits overlay. M kuhner (talk) 15:19, 2 June 2026 (UTC)[]
• Support per above Fortek67 (talk) 15:26, 2 June 2026 (UTC)[]
• Support for the reasons that others above have articulated. Stepwise Continuous Dysfunction (talk) 20:41, 2 June 2026 (UTC)[]

I don't think you need any particular extra consensus, as they are blocked for AI use, which they admitted to, and have not come back to argue for unblocking. --Joy (talk) 21:38, 7 June 2026 (UTC)[]

Under WP:LLMPRV § Requirements 1A, the user needs to have been blocked for at least a week and have no active appeals for LLMPRV to be used. But by gaining consensus here, we can use 1D instead and get started right away. And having consensus makes potentially controversial actions easier in general. ‑‑gurkubondinn 21:53, 7 June 2026 (UTC)[]

It's been about a week already. WP:Wikipedia is not a bureaucracy, we don't have to measure minutes. --Joy (talk) 10:34, 8 June 2026 (UTC)[]

But it hadn't been a week when the consensus was gathered. It just makes things easier when you do it this way. Wikipedia isn't a bureaucracy indeed, but it is very easy for time consuming cleanup tasks to get bogged down in various ways. A few support !votes like this can be a good way to prevent that, and LLMPRV has been a good step in the right direction. ‑‑gurkubondinn 10:43, 8 June 2026 (UTC)[]

Agreed on Srebrenica, so I've rolled the edits back. Please check if I missed something, this is definitely an article that deserves every effort to make sure the Slop Machine™ 3000 has not touched it. Or at least it's not something I'm willing to take any chances on. --gurkubondinn 11:47, 2 June 2026 (UTC)[]

Quick checked Srebrenica, thanks for doing that, and it looks OK. Indeed it was only after your reversion that I could see how LLM had elected to downplay and dumb down the events of 1995. ChrysGalley (talk) 11:57, 2 June 2026 (UTC)[]

Well, yikes.

• AI: "During the Bosnian War, Srebrenica became internationally known as the site of the Srebrenica massacre,... The massacre was later ruled to be an act of genocide...[emphasis added]"
• Restored: During the Bosnian War in 1995, Srebrenica was the site of genocidal killing..., which was subsequently designated as an act of genocide...[emphasis added]

This is a bit subtle, but it unequivocally downplays the events of 1995. Thanks for pointing this out, I hadn't noticed that they had edited the Srebrenica article. --gurkubondinn 12:14, 2 June 2026 (UTC)[]

• EiersalatmitGurken (talk · contribs)

Numerous editors have tried raising concerns on their talk page, only be reverted without any real answers: History/User talk:EiersalatmitGurken. They have however given some answers on User talk:Belbury § Stop posting on my talk page, and then also came to my talk page to post User talk:gurkubondinn § Stop posting on my talk page.

They've also made an AI-generated edit request on Talk:Iran § Update request for the Religion section (Secular shift, mosque closures, and growth of Christianity), which I think is worrying enough on its own. The references that they gave just pointed to the landing pages of the websites and did not link to the articles that they wanted to have referenced in the article (though they do appear to exist this suggests that they haven't actually read them). Some of their translations seem to have been unattributed, and they have conceded that these are machine translations so they most likely warrant some closer scrutiny, along with the rest of their contributions. --gurkubondinn 17:04, 1 June 2026 (UTC)[]

They've also been adding WP:AIIMAGES to articles, apparently these have been made with Grok: User talk:Fermiboson § Please do not write on my discussion page again. --gurkubondinn 17:11, 1 June 2026 (UTC)[]

The article Walter Magaya met the G15 criteria when it was created: Diff/1294229374. There are 2x sources that return 404 not found (just linking to www.bbc.com/news/world-africa) and 1x just links to the a landing page (www.herald.co.zw). Both of them even remain in the current revision of the article. --gurkubondinn 17:50, 1 June 2026 (UTC)[]

Went through the reflist of Walter Magaya (at revision 1345449859):

• Refs nr. 1, nr. 17, nr. 18, nr. 19, nr. 23, and nr. 24, return 404 not found.
• Refs nr. 2, nr. 6, and nr. 8 is the BBC source that was present on creation and returns 404 not found (www.bbc.com/news/world-africa), with a couple of different titles.
• Ref nr. 5 is just points to the front page of the Herald (Zimbabwe), present since creation (www.herald.co.zw)
• Refs nr. 20, nr. 21, and nr. 30 return 500 server error, but the main page (www.newsday.co.zw) appears to load fine.

This article might still meet the G15 criteria.. --gurkubondinn 18:55, 1 June 2026 (UTC)[]

I've gone and started WP:AfD/Walter Magaya for this article. --gurkubondinn 09:25, 2 June 2026 (UTC)[]

I left a notice on their talk page about this thread. [60] Kodning 🌸 (talk) 20:11, 1 June 2026 (UTC)[]

They'll probably revert you. I figured it would probably be more productive to inform them in the conversation on my own talk page instead: Diff/1357256371. --gurkubondinn 20:17, 1 June 2026 (UTC)[]

They're continuing, and just inserted this edit which they claim is "translated with AI but reviewed and corrected": Diff/1357284108. This referenced a paper titled "Causes of longitudinal nail splitting: a retrospective 56-case series", but it was referenced with with PMID 35080060 which is assigned to an unrelated paper "Heart rate variability increases following automated acoustic slow wave sleep enhancement". The DOI was correct though, but I have reverted it regardless since it was confirmed as AI-generated in the edit summary, and the incorrect PMID shows that it does fails the WP:LLMT guideline. --gurkubondinn 20:31, 1 June 2026 (UTC)[]

And now they have restored this AI-generated edit (but removing the hallucinated PMID). This continued use of AI is starting to feel disruptive at this point, but I won't be reverting again on this article. --gurkubondinn 20:57, 1 June 2026 (UTC)[]

I was trying to save time and didn't want to write the entire reference manually, so I just provided the link to the reference and asked the AI to create the citation from it. I reviewed it several times, but unfortunately I missed checking the PMID. Going forward, I will verify every parameter individually.

My apologies. I underestimated how prone AI is to hallucinations in such details. Thank you for catching this. EiersalatmitGurken (talk) 20:58, 1 June 2026 (UTC)[]

...,so I just provided the link to the reference and asked the AI to create the citation from it.

This is not allowed per WP:NOLLM and is not at all how you have claimed to use LLMs so far. Together with the other sourcing related problems that I have identified in your edits, I think that you should pause editing article space until this has been cleared up. It is simply not feasible to add more to the pile for others to go through and validate or clean up. And quite frankly, I am starting to be worried that you may be getting yourself blocked. --gurkubondinn 21:12, 1 June 2026 (UTC)[]

I wasn't aware how bad the hallucination is. I realized that now. Thank you for making Wikipedia a place with high quality. That's why I love Wikipedia. EiersalatmitGurken (talk) 21:37, 1 June 2026 (UTC)[]

Here is a list of tools that would be better for expanding citations: WP:CITEGENERATORS Kodning 🌸 (talk) 21:54, 1 June 2026 (UTC)[]

You can also read this essay that I wrote recently after explaining the same to another user:

• User:gurkubondinn/REFPRIMER: a primer on referencing

--gurkubondinn 21:57, 1 June 2026 (UTC)[]

Thank you very much! :) EiersalatmitGurken (talk) 22:05, 1 June 2026 (UTC)[]

Thanks a lot! EiersalatmitGurken (talk) 22:05, 1 June 2026 (UTC)[]

LLM content on medical topics is a particularly bad idea (even worse than all the other places where it is already bad). Medicine is where we have to be particularly careful. Stepwise Continuous Dysfunction (talk) 21:02, 2 June 2026 (UTC)[]

Warned per evidence at this thread. If any further warnings are removed by this user, we may need to fall back on WP:AIV. | One Reaction was here. Got a complaint? 20:58, 1 June 2026 (UTC)[]

It would not be the removal of warnings (which is explicitly permitted) that would be the issue, continued editing with LLMs is the issue. I'm not sure the user understands that their use is not allowed. @EiersalatmitGurken, could you please clearly and explicitly commit to ceasing any use of LLMs in the course of your Wikipedia editing in the future? Fermiboson (talk) 22:24, 1 June 2026 (UTC)[]

Thanks for the information, didn't get that I'm allowed to edit my discussion page. I will use LLMs only the way it is allowed, aka only in translations if they are very well checked and for basic copyedits. EiersalatmitGurken (talk) 22:43, 1 June 2026 (UTC)[]

You are allowed to remove most things from your talk page, the list of things that you are not allowed to remove is very short. But just because it is allowed does not mean that it is a good idea. Removing legitimate concerns from other editors and outright refusing to discuss your own conduct and editing on your own talk page is will at worst be interpreted as evidence that you are incompatible with working on a collaborative project, and will not work in your favour. This is not grounds for sanctions, but if yuo would find yourself in a situation where someone is evaluating weather or not to sanction you over something else, this behaviour would not work in your favour. Please take this as the friendly advice that it is meant as.

Now about your AI use; I think it would be for the better if you took @Fermiboson's advice and committed to completely ceasing any and all use of AI on the English Wikipedia going forward. This has already created numerous problems, some of which violate our verifiability policy, which might just be our most important policy around this place. And my own advice to you would be to stop making edits in article space for now, and spend your time going through your past edits and 1) removing anything that was generated by an AI and 2) reading and verifying every single citation that you have added anywhere. These two things would go along way towards convincing your fellow volunteer editors that you have the required competency to edit here. --gurkubondinn 22:58, 1 June 2026 (UTC)[]

Thanks. I will take your advice and review my past edits and improve if needed - before making new ones. Verifiability is extremely important. I agree with you 100%.

About my talk page I have a different opinion. It is my space and I can use it as I want. Concluding that somebody doesn't want to collaborate just because he prefers to control his discussion page is wrong. Just inform the user about the issue on the relevant page, like here. EiersalatmitGurken (talk) 23:25, 1 June 2026 (UTC)[]

You don't "control" your talk page though. You are given a wide berth when it comes to your userspace, but you still don't WP:OWN it. Refusing to discuss your own conduct and editing is being uncollaborative because you are forcing the discussion to happen on a noticeboard where it takes up the time of more editors. Granted, an uncollaborative editor will eventually end up on a noticeboard regardless of how they treat their talk page, but that's just furthering the point. I would urge you to adopt a more collaborative attitude towards other editors here. --gurkubondinn 10:15, 2 June 2026 (UTC)[]

Kyleoahso (talk · contribs · deleted contribs · logs · filter log · block user · block log)

[61] "Got you bro." WP:DUCK Fermiboson (talk) 22:21, 1 June 2026 (UTC)[]

Reverted to last non-LLM version in April. Sarsenet•he/they•(talk) 23:48, 1 June 2026 (UTC)[]

E43-VrGa (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Blatantly obvious LLM use. G15'd Dolores Vázquez for hallucinated references. New, 1600-word, version created just 19 minutes later with multiple nonexistent reference URLs. ~ A412 ^talk! 16:15, 2 June 2026 (UTC)[]

Requested G15 for Nguyễn Thiên Phụng over similar referencing. --gurkubondinn 19:50, 2 June 2026 (UTC)[]

And for John S. Bradstreet as well, same deal. --gurkubondinn 20:04, 2 June 2026 (UTC)[]

Requested G15 for Prince Group transnational criminal organization case--it's a shame, the topic is unquestionably important and significant, but the sourcing is a hallucinatory nightmare. I left a detailed breakdown on the Talk page. M kuhner (talk) 06:22, 6 June 2026 (UTC)[]

The way I see it, deleting LLM-generated pages isn't about removing information about the topic from the encyclopaedia, it's just about removing LLM-generated text. Deleting the page itself, not the topic.

And yes, I know this is WP:AIPARALLEL § Not X, but Y but I can't be bothered to think of a different way to write it. --gurkubondinn 06:30, 6 June 2026 (UTC)[]

Indeed. But it would be a lot of work to corral that many real sources (a decent number are real) and write a good article about them: so it's a shame this one is not that article.

I do think there is some tension in the AfD rules, where generally speaking we don't delete an article that could be saved, but for LLM the consensus seems to be delete anyway. Having tried my hand at fixing LLM articles, I now see why. Expanding a clean article is so much easier than rewriting an LLM article to be actually correct. Something like Prince Group looks like a good start, but actually it's a nasty trap. M kuhner (talk) 07:04, 6 June 2026 (UTC)[]

I think there's three elements here. As you observed, "expanding" an LLM-generated start to an article is at least as hard as writing the article from scratch. TNT stubify is a more common result with other content problems (copyvio comes to mind) where it's possible to separate the problematic content, but with LLM articles there is usually no content we can confirm to be nonproblematic and reliably sourced. Third, I think there's elements of WP:DENY and protecting the project that influence !voters' minds. ~ A412 ^talk! 15:12, 6 June 2026 (UTC)[]

Try using WP:TWINKLE for stuff like G15. You've done them "by hand" a couple of times now, so you understand what's involved. ‑‑gurkubondinn 07:32, 6 June 2026 (UTC)[]

Thanks, will do. It's after midnight here and I'm not as bright as I should be.... M kuhner (talk) 07:40, 6 June 2026 (UTC)[]

My G15 deletion of Prince Group transnational criminal organization case was declined by an editor who pointed out that the article is translated from the Chinese Wikipedia, and that most of the anomalies (article titles, dates, non-exact quotes) could be due to translation rather than LLM. On examination they have a point. On the other hand, early spot-checks of source/text made me very uneasy. I'm going to do a detailed source/text analysis today if I can. I'm wondering, though, about how to evaluate a translation. It's not that surprising that a quote in an English source, translated to Chinese to be direct-quoted in the Chinese article, and translated back to English to be direct-quoted in the English article, would be far from correct on arrival. But presumably it's the translator's job to fix that. M kuhner (talk) 00:09, 7 June 2026 (UTC)[]

I put a source/text check for the first half of Prince Group transnational criminal organization case on its Talk page. I was only able to check English language sources. Executive summary: the vast majority of claims failed to be supported by their source. Many, perhaps most, could have been supported by sources used elsewhere in the article, though at least one didn't match any source I read. It is a complete trainwreck.

I could use advice on what to do next. AfD? The article is unquestionably notable, though. Some kind of stubification? @Gurkubondinn, @A412, any suggestions?

Really sad about this one. It's an article that needs to exist; it deserves better than this. M kuhner (talk) 01:41, 7 June 2026 (UTC)[]

I actually think that stubification would be ideal, since this is a pretty important topic. But if it doesn't work or the AI-generated text gets reinstated, then I think AfD would be warrented. TNT will be required in some form or another though, since AI-generated articles can't really easily be salvaged without rewriting anyway. Stubification would retain the sources in the page history, which might make it easier if someone wants to rewrite this at some point. ‑‑gurkubondinn 15:58, 7 June 2026 (UTC)[]

The last article, Prince Group transnational criminal organization case, is stubified and at AfD; one way or another I think it's dealt with. Can this be closed? M kuhner (talk) 03:42, 9 June 2026 (UTC)[]

Agree this can be closed. What's the closure procedure? ~ A412 ^talk! 22:21, 9 June 2026 (UTC)[]

Maybe not ready for close yet. Their Mihovil Abramić just got passed through AFC. ~ A412 ^talk! 02:48, 10 June 2026 (UTC)[]

Phooey. Most sources are in Croatian, but In addition to his fieldwork, Abramić served as the primary editor of the Vjesnik za arheologiju i historiju dalmatinsku (Bulletin of Dalmatian Archaeology and History) during his tenure in Split, curating decades of regional historical research. is sourced to a page whose only mention of Abramic is It consists of the bequests of Don Frano Bulić, Mihovil Abramić, Luka Jelić, Francesco Carrara, Julije Bajamonti and the Pavlović-Lučić family.. Same old problem. I think we need admin attention NOW. M kuhner (talk) 03:10, 10 June 2026 (UTC)[]

@Admins willing to patrol AINB: Requesting some form of block to prevent User:E43-VrGa from creating more LLM pages. They have been warned and have had previous articles deleted for LLM use, but have just created another article without responding to any of this. That article, Mihovil Abramić, has source/text mismatch and we believe it to be LLM as well. M kuhner (talk) 03:21, 10 June 2026 (UTC)[]

These articles appear to have been created earlier (back in May) and more recently accepted through AfC. Looking at their contributions, I don't see anything from after they received the LLM warnings. Chaotic Enby (in solidarity · talk · contribs) 07:39, 10 June 2026 (UTC)[]

You're right: I didn't think of that. Apologies. M kuhner (talk) 14:59, 10 June 2026 (UTC)[]

This user appears to have returned after a long wikibreak to add a lot of AI-generated content to various articles. (i.e., this obviously does not apply to anything from before 2026)

I have left a warning for this; others have warned them for non-reliable sourcing and what appears to be a possible hallucination, both also characteristic of AI problems. Gnomingstuff (talk) 15:02, 4 June 2026 (UTC)[]

Also on their Talk page: Thanks for reviewing the updates. For what it's worth, my use of the term wasn't stemming from YouTube, but rather from the general drift of the term in broad English-language pedagogy. That said, I completely agree with you that strictly adhering to precise phonological typologies makes for a much stronger encyclopedia article. Sounds nothing like their previous conversations (pre-2026). M kuhner (talk) 17:45, 4 June 2026 (UTC)[]

I wouldn't necessarily place too much weight on difference in writing styles given that it's been over a decade, but their communication on Talk:Otaku is not encouraging. Gnomingstuff (talk) 22:13, 5 June 2026 (UTC)[]

Not just difference in writing styles, though: it's challenging to believe that a human wrote the bit I quoted. "Broad English-language pedagogy" -- a human could be that pompous, but a human being that pompous wouldn't say "pedagogy" (educational praxis) but instead "scholarship" (scholarly praxis) or something similar. Also "completely" "strictly" "precise" and "much" all in one sentence.

Talk:Otaku, though, may make that completely moot. They say they used LLM for citation formatting, but then offer to go through and de-LLM the article text: would you be open to me restoring the core academic citations and structure, provided I manually rewrite the prose to ensure it meets a standard, human-crafted encyclopedic tone? This strikes me as a tacit admission of LLM text. (And is LLM itself.)

Are we at the point of flagging for an admin? M kuhner (talk) 22:35, 5 June 2026 (UTC)[]

I mean I'd rather they weigh in first especially since they were editing over a decade ago Gnomingstuff (talk) 00:15, 6 June 2026 (UTC)[]

They have made another fairly large edit as of June 7. I spot-checked a citation from it. Minkova, Donka (2014). A Historical Phonology of English. Edinburgh University Press. pp. 143–144. ISBN 978-0748634699. The ISBN does lead to that book, but Google Books says 2013 not 2104, and "Historical Phonology of English" not "A Historical Phonology of English." I don't know if that's a concerning difference or not. The article is so extremely technical that I can't otherwise evaluate it. M kuhner (talk) 03:52, 9 June 2026 (UTC)[]

They're also restoring content Gnoming removed, while ignoring questions about AI use on the talk page [62]. They also restored this, which is practically all uncited, except for a paragraph of prose (starts A similar contrast is seen with the imperative form of ...) cited to an index. @Admins willing to patrol AINB: idk whether a mainspace block would be good here Kowal2701 (talk, contribs) 10:53, 13 June 2026 (UTC)[]

I left a final warning to refrain from making further substantive changes without addressing the concerns raised. Ping me if further action is required. ~Anachronist (who / me) (talk) 16:50, 13 June 2026 (UTC)[]

Also see WP:AI noticeboard/Archive 5 § Preye douglas. This user was recently temp blocked from mainspace [63] for repeated AI misuse by Dr vulpes. They responded by denying obvious LLM use and complaining about the block [64][65], waited out the block, then cranked out Woman Commando, which is mostly AI-generated. The article treats us to prose such as Lyrically, "Woman Commando" is a women's empowerment anthem centered on themes of feminine confidence, luxury and revelry. NicheSports (talk) 07:05, 5 June 2026 (UTC)[]

I am not sure. I clicked into every source: 100% live sources that go to what they say they do. I spot-checked accuracy of the text in a handful: found nothing. I was sure that Meanwhile, Maduafokwa concluded that the song "creates an argument about cross-continental connection [...] placing womanhood at its thematic center." would turn out to be bogus, but no, that's exactly what the source says!

I recommend wait and see. In my experience LLM articles never have sourcing this good, though it could be diligent hand correction of initially LLM text. M kuhner (talk) 17:13, 5 June 2026 (UTC)[]

Hi M kuhner, please compare the style of writing to that of the user's comments, for example [66] at the previous thread. They did not write this article (which is exactly the type of article they were blocked for - also see Beggie Beggie, Aye Kan (Are You Coming Back) etc.) without LLM assistance. NicheSports (talk) 17:32, 5 June 2026 (UTC)[]

In my experience LLM articles never have sourcing this good,

Direct quotes tend to be accurate with modern chatbots, hallucinated quotes are rare unless it's rewriting an existing quote. Where LLMs fuck up is any kind of interpretation of that text. Gnomingstuff (talk) 22:14, 5 June 2026 (UTC)[]

It's the 23/23 live sources with correct titles and relevant content that gives me pause. I tend to hit about 1/3 to 1/4 bad or dead sources in LLM articles. If it is LLM, it has been hand-edited for source validity. If I have time this evening I'll try a detailed check for source/text correspondence, but it will take a long time.

I wasn't putting forward the direct quote as particularly evidentiary--I agree that they get interpretation wrong more often than quotes--but just to express startlement that it was correctly sourced, as it sounds so bad! M kuhner (talk) 22:26, 5 June 2026 (UTC)[]

I think this is probably a case of partially rewriting/polishing with AI and/or using AI to summarize sources; there's a lot of stuff in here that doesn't seem grammatically/syntactically like raw AI output at all. I don't think it's the other way around ("hand correcting" AI output) because if this was originally AI output you'd have to really, really know what you were doing at an in-the-weeds syntax level to "correct" it in that way (e.g. to insert misplaced modifiers, tiny prepositional phrases, awkward constructions, etc). Gnomingstuff (talk) 00:22, 6 June 2026 (UTC)[]

Agreed that not all of it is AI. What are your thoughts on this source-to-text integrity analysis of the Composition section, which is the one that gave the strongest signs of AI-generated text (this is every sentence in that section btw, so no cherry-picking)

• "Woman Commando" runs for three minutes and twenty-eight seconds. Set over a tempo of 115 beats per minute, the song is written in the key of B minor and follows a chord progression of B minor, F♯ minor, E, and E minor. mostly verifies to [67], but that source is purely reference (has no written info). As a result, the bolded bit might be OR, and implying that E minor comes last in a "progression" is incorrect, as (from my review of the source) the primary progression is B minor, F# minor, and E. E minor is a bridge following B minor in some parts of the song
• It is a multilingual afrobeats and amapiano track that fuses latin pop and R&B influences, incorporating English, Nigerian Pidgin, and Portuguese. 👁 Red X
  N bolded fails verification in either source [68][69]
• Starr described the production as driven by log drums, which Kyle Denis of Billboard said gave the song its "urgency and irresistibility" 👁 Red X
  N bolded bit fails in the provided source [70]
• Blossom Maduafokwa, writing for The Native, deemed the amapiano arrangement "club-ready" 👁 check
  Y verifies in [71]
• while Quincy from Ratings Game Music felt that the song carries a lively tempo, dramatic violins, and an array of melodies this verifies in the source [72] but is a clear case of WP:CLOP: "lively tempo" "dramatic violins" and "array of [fun] melodies" are all pulled straight from the source
• Lyrically, "Woman Commando" is a women's empowerment anthem centered on themes of feminine confidence, luxury and revelry. 👁 Red X
  N I don't think this verifies in either source. It certainly fails in [73]. The only bit in [74] that is close is and lyricism about popping Hennessy at tables - but going from that to "luxury and revelry" gives off strong vibes of LLM SYNTH
• In the first verse, Starr encourages solidarity with "Nobody's gon' be left behind / As a lioness I dey move with squad", before opening the second verse with "Tonight e be ladies night / I no wan know your zodiac sign." 👁 Red X
  N bolded bits (most of this) fail verification in the source [75]
• Anitta's verse includes a lyrical reference to "1942", another track on The Year I Turned 21, with the lines "Má' botellas llegando / 1942, tomando." 👁 check
  Y verifies in the source [76]
• Denis praised the vocal chemistry between the three artists, observing that Jones's "rich tone" pairs well with Starr's and that Anitta's delivery "intensifies the track's swagger." 👁 check
  Y verifies in source [77]
• Meanwhile, Maduafokwa concluded that the song "creates an argument about cross-continental connection [...] placing womanhood at its thematic center." 👁 check
  Y verifies in source [78]

So by my count, of the 10 sentences in this section, which has a different style than that of Preye douglas' comments, we have

• 4 sentences that clearly verify
• 4 sentences with clear verification failures
• 1 sentence that mostly verifies but may have some OR / minor issue with chord progression
• 1 sentence of clear CLOP

What do you think? NicheSports (talk) 02:31, 6 June 2026 (UTC)[]

Yeah this was basically my take, it feels like AI source summarization Gnomingstuff (talk) 03:05, 6 June 2026 (UTC)[]

What do you mean by AI source summarization? NicheSports (talk) 03:14, 6 June 2026 (UTC)[]

using AI to "summarize" a source Gnomingstuff (talk) 03:35, 6 June 2026 (UTC)[]

I withdraw my objection. Thanks for doing the detailed analysis! M kuhner (talk) 06:24, 6 June 2026 (UTC)[]

I totally disagree. This was 100% written without AI. It was written by me with help and inspiration from previous Wikipedia articles and according to what I read from sources provided. Multiple sources including Billboard confirmed the song as being multilingual. Starr said in an interview that she wanted Anitta on the track because she sings in Portuguese and will bring a Latin vibe to it and normally the Starr's lyrics were in Pidgin and Coco Jones sang mostly in English. In her Apple Music album notes, Starr described "Woman Commando" as "such an Afrobeats/amapiano banger in a way, because of the log drums." Furthermore, in the Billboard review of TYIT21, Kyle Denis said, "On the latter, a multi-lingual collaboration with fellow Grammy nominees Coco Jones and Anitta, Ayra delivers a women-empowerment anthem (“Tonight e be ladies night/ I no wan know your zodiac sign”) that sources its urgency and irresistibility from those pounding log drums. Big-name crossover collaborations tend to collapse under their own weight, but Ayra’s ear for vocal chemistry – Coco’s rich tone pairs beautifully with hers and Anitta’s cavalier delivery only intensifies the track’s swagger – is particularly special. Of course, all these hymns of independence and confidence exist in conversation with songs exploring the darker parts of Ayra’s early 20s." Preye douglas (talk) 02:42, 11 June 2026 (UTC)[]

Rather than telling us about the song, could you please address the sourcing issues raised by NicheSports above, specifically the four marked with red x-marks? How did those happen? How, specifically, are you going to change your article writing strategy to stop them happening again?

It is not enough for a claim to be true. If we say "Source A says X," source A must say X. If it's in Source B instead, which is a very, very common LLM mistake, that's not acceptable. Claiming people said something they didn't actually say--and those four red x-marks are exactly that--is no good. M kuhner (talk) 02:59, 11 June 2026 (UTC)[]

Kyle Denis was the writer of the billboard review so therefore he said it, I only paraphrase a little bit but I made sure the important parts of his opinions were in quotes. Same goes for Starr, check the Apple Music album notes. A lot of Wikipedia articles on songs do the same thing, paraphrase and add the writer's quote instead of copypasting the entire as long as what was paraphrased isn't far from the intended meaning. Preye douglas (talk) 03:07, 11 June 2026 (UTC)[]

Last try.

Up above is a quote from the article that says the song is multilingual and in English, Nigerian Pidgin, and Portuguese. Your article cites this to two sources. Neither one says that it is multilingual, nor that it's in English, Nigerian Pidgin, and Portuguese. I just read both from end to end. It's just not in there. You don't get to claim sources say things they don't actually say. It does not matter if those things are true (I believe this one is true). You cannot attribute it to a source that does not say it!

This is fundamental. If you can't learn this, you can't edit Wikipedia. M kuhner (talk) 03:16, 11 June 2026 (UTC)[]

This goes to show that you ignored reading my response above. The billboard review by Kyle Denis clear says "On the latter, a multi-lingual collaboration with fellow Grammy nominees Coco Jones and Anitta, Ayra delivers a women-empowerment anthem" and Sandra Appiah from Face2FaceAfrica also said, she diversifies her Afropop/R&B palate by exploring other sounds, as Amapiano’s rollicking log drums power the multilingual female empowerment anthem Woman Commando. Preye douglas (talk) 03:30, 11 June 2026 (UTC)[]

Neither of the links given in the article to source that statement go to the billboard review by Kyle Denis. One goes to Shaun Grant's article. The other goes to Sandra Appiah's article. The statement is true but the claim that it is in these sources is false. You needed to source it to the Billboard review, and you didn't. M kuhner (talk) 03:38, 11 June 2026 (UTC)[]

so what is Reference #10? Preye douglas (talk) 04:13, 11 June 2026 (UTC)[]

Reference #10 is the Billboard review. It should have been cited as the source for the language claims, but it is not, as you can clearly see in the article. They are cited to Grant and Appiah, who didn't say that. M kuhner (talk) 04:16, 11 June 2026 (UTC)[]

Although the lyrics were in Pidgin English it was basically Starr bragging about living a luxury lifestyle and partying with her fellow girls which is the same revelry.

All my G's fit to testify (yeah)

Hennessy when we gettin' high

At the scene 3:39

Tell the Mercedes to park inside

Girls plenty for my table side

Another set still dey come from house

Nobody's gon' be left behind

As a lioness I dey move with squad Preye douglas (talk) 03:00, 11 June 2026 (UTC)[]

Any interpretation of lyrics must be cited to a source. Gnomingstuff (talk) 05:47, 11 June 2026 (UTC)[]

Of course and I did include a source to support that. See reference #13. Writer Thomas Melia states: Starr asks us to “Follow the woman commando”, on the female-fronted fourth track, with encouraging lyrics like “Nobody’s gon’ be left behind / As a lioness I dey move with squad”.

Since Melia describes the lyrics as "encouraging" and the lines themselves clearly convey solidarity, the combination is reasonable. Preye douglas (talk) 09:40, 11 June 2026 (UTC)[]

Looking back through some recent edits and noticed [79] which they then immediately reverted [80]. It contained this The Year I Turned 21 adopts a more introspective tone than her previous work, navigating themes of self-discovery and independence. which was sourced only to the promotional [81] and completely failed verification. Was this really written without AI assistance? NicheSports (talk) 20:55, 11 June 2026 (UTC)[]

First of all, you are wrong to call this a recent edit because it wasn't. Many editors have worked on that section since then. While trying to do a clean up which I doubt was necessary, one editor mistakenly removed the citations that sourced that statement, but since you have chosen to bring it up, let me show you how I came up with it:

1. Kyle Denis in his Billboard review wrote that her debut 19 & Dangerous "presented a self-assured teenage star who effortlessly and thoughtfully navigated the transition from adolescence to young adulthood."

2. Robin Murray from Clash Magazine also said she can twist from club shakers to sombre moments of reflection, with her voice guiding you to a place of hushed introspection."

3. Apple Music's editorial describes the album as "brimming with self-confidence and "unvarnished introspection"

4. Ayra Starr herself in an interview described the album as "a coming-of-age story" about "growth, realizing your power, and love" and called it an autobiography touching on themes of self-discovery. Preye douglas (talk) 04:02, 12 June 2026 (UTC)[]

Hi @Preye douglas, claims must have a reliable source directly attached to the statement now an inline citation so that the reader can verify what you've written. This was new text that you wrote and cited to a specific source - nobody else wrote that text that I can see and you didn't give any of those sources at the time.

Citations are useless if they're not in the right place or not in the article at all, since the reader has no way to fact check what you've written.

We don't need to know what sources match the claim now, we need to know why you made this edit at the time.

To be clear, can you please confirm:

• Why did you choose the pulse.ng source to verify your claim that The Year I Turned 21 adopts a more introspective tone than her previous work, navigating themes of self-discovery and independence?

In solidarity, Blue-Sonnet (I'm listening) 06:58, 12 June 2026 (UTC)[]

Just to be clear, if I write Dave Smith likes chips I must add a reliable source at the end of that specific sentence, so the reader can confirm that Dave does indeed like chips.

The source cannot be on a different part of the article or left out of the article - that's not how Wikipedia works. It's also really confusing and not fair to the reader, because we don't expect them to do their own verification of what we write. That's our job as editors.

Please read through Wikipedia:Verifiability to see what I mean - if you add a claim, you must always add an inline citation that backs up that specific claim, and it must be done when you add the text (or very shortly afterwards).

If we didn't have this safety measure of claim verification, anybody could write whatever they liked and not have to provide any proof. A citation that isn't given in the text is useless to the reader, and we write articles specifically for the benefit of the reader.

With this in mind, please let me know the answer to my question above after you've read up on Wikipedias Verifiability policy. In solidarity, Blue-Sonnet (I'm listening) 07:22, 12 June 2026 (UTC)[]

Thank you for the clarification @Blue-Sonnet. You are right that the inline citation needs to be directly attached to the statement, and I accept that. To be transparent about what happened: during a random cleanup by another editor, the citations that were supporting that statement got removed while the Pulse Nigeria reference was left behind even though it was meant for a different claim entirely. That is what created the mismatch you are seeing now. Preye douglas (talk) 09:06, 12 June 2026 (UTC)[]

@Preye douglas this was not a random cleanup but one of a half dozen editors [82][83][84][85][86][87][88] who wrote in their edit summaries that they were addressing issues with AI-generated content in this article. The text in question appears to have been originally added by you at [89] on 17 October 2024, at a time by your own admission you were using AI assistance to edit articles. Note: when you originally added this statement it was sourced to [90] which also fails verification. So at [91] you restored what appears to be 2024-era AI-generated content, verbatim, and randomly (?) selected a source to back up the AI-generated statement. You did revert this yourself, which is good. But if you can't be honest about the process and where the text came from, how can we trust that you are being honest in answering questions about potential AI usage on other recent edits such as your creation of Woman Commando? Hopefully you see why editors here are expressing concern about your editing approach. An answer like "I was restoring some content that had been removed, then reverted myself when I realized it included AI-generated text from an earlier edit" would be a totally fair answer btw. NicheSports (talk) 20:13, 12 June 2026 (UTC)[]

I have been honest from the start it's just that you seem to like to cling to the past. I have not been using AI help for recent edits especially after the first time you and bink' brought up the issue several months back. It's true that I did get some LLM assistance in the past to save time but not anymore. And it wasn't even prohibited at the time. Preye douglas (talk) 20:35, 12 June 2026 (UTC)[]

I have tagged Woman Commando as LLM. It has minimal content from any other editor: should it be deleted? M kuhner (talk) 03:31, 10 June 2026 (UTC)[]

Hi @M kuhner, I do not think there is consensus on this thread about the degree to which AI was used to assist with that article, so I would would off on deletion unless/until that consensus emerges. Tagging the Composition section is probably fair though given the analysis above NicheSports (talk) 03:38, 10 June 2026 (UTC)[]

Appears to be using LLMs to summarize/paraphrase sources but has denied LLM usage on their talk page. Akani (Arcania) for example shows WP:AIATTR with repeated mentions of "accounts" and "reports" and "sources", and WP:AITREND with reflecting the growing scale of inland gold trade with the coast. Draft:Akanman Piesie Nnum has the same issues, e.g. more WP:AITREND with Other traditions place these movements within broader tensions in Adanse and Amansie. When I drilled into material in Draft:Akanman Piesie Nnum cited to Kea 2000, pp. 522–523, the draft's claim suggesting that the town served important religious and social functions is a hallucination. On List of rulers of the Akan state of Bono-Tekyiman, we have List of Rulers of the Akan state of Bono-Tekyiman outlines the traditional succession of kings and queenmothers....

Other articles that may need to be cleaned up include Bono State, Asante Empire, Akwamu Empire, Kingdom of Gyaman, Adanse, Akan clans, Akan people, Ghana, History of Ghana, Kingdom of Assin, Kingdom of Denkyira, Kingdom of Twifo, Adansemanso, Wenchi, Fetu Kingdom, Kingdom of Aowin, Begho, Old Banda. I haven't notified Oti Ne Asare of this discussion. Helpful Raccoon (talk) 21:01, 6 June 2026 (UTC)[]

I've declined the draft as AI generated based on your assessment here. No time to dive into the other articles for now. Athanelar (talk) 07:21, 7 June 2026 (UTC)[]

Asantemanso was created just 22 minutes after Adansemanso... Asantemanso at the time contained a red-linked category, Cultural heritage monuments in Ghana. The section Comparative sites has signs of LLM language: Nearby sites excavated for comparison include: The word conquest at Adansemanso#Decline and Conquest has an unnecessary capital letter, which was later corrected. Fortek67 (talk) 09:52, 7 June 2026 (UTC)[]

Old Banda, Fetu Kingdom and Kingdom of Assin has clear usage of AI (indicated by the astericks used to bolden or italicize texts). Kingdom of Aowin then has a lead which seems very artificially generated. The whole general references list at the bottom would also point to possible usage of AI. Speaking of that, sources and references are separate sections?

PeepeeDino (talk) 12:14, 7 June 2026 (UTC)[]

Yep. That confirms it more. All articles created by Oti Ne Asare have been tagged for potential AI use. I've tagged some other articles edited by the user too. I think it's now appropriate to nominate all articles for deletion and revert LLM content per WP:LLMPRV. Should I go ahead? Fortek67 (talk) 12:18, 7 June 2026 (UTC)[]

Preferable but maybe wait a bit for more consensus on this? Just to have a wider agreement I suppose. Oh and also maybe putting the case in the noticeboard and their talk page. See what they have to say.

PeepeeDino (talk) 12:30, 7 June 2026 (UTC)[]

Yeah I meant consensus sorry Fortek67 (talk) 15:27, 7 June 2026 (UTC)[]

No, none of the conditions at WP:LLMPRV are satisfied. I’ve explained and asked them to comment here. Tbh I’ve interacted w Oti a fair bit, they’re very knowledgeable and competent w the topics, I hope this can be resolved by them doing the cleanup Kowal2701 (talk, contribs) 14:02, 7 June 2026 (UTC)[]

So perhaps go for draftification if the editor is willing to fix the articles? I thought it would have been alright per this policy 3. The contributions to be removed coincide in timing and pattern with edits determined to very likely be AI-generated. Fortek67 (talk) 15:28, 7 June 2026 (UTC)[]

Speaking of that, sources and references are separate sections? This seems to be an emerging new AISIGN. I've seen it in a couple of drafts in the past few days. Athanelar (talk) 13:27, 7 June 2026 (UTC)[]

It sure is, seen it too. Fortek67 (talk) 13:32, 7 June 2026 (UTC)[]

Goodafternoon, thanks for notifying me. Some of these articles were created or expanded when I was a new editor. I did use grammarly and other writing tools like synonym finder to help with wording, and grammatical errors. Some editors year flagged old edits I’ve since been trying correct for some time now. Because I made a large number of edits when I first started, I’m still in the process of correcting/cleaning them. I’ll reread the policies again and revisit every article I contributed to when I first started to fix everything brought up here if that’s possible. Oti Ne Asare (talk) 17:06, 7 June 2026 (UTC)[]

Thanks for your understanding, Oti. I think it'll be better for me to go ahead with draftification of all articles instead of mass deletion, this will give you the chance to improve and fix the articles. Fortek67 (talk) 17:22, 7 June 2026 (UTC)[]

All seven articles created by Oti have been drafted. Fortek67 (talk) 21:15, 7 June 2026 (UTC)[]

copying over from the talk page (@Oti Ne Asare)

Hey, thanks for bringing this up because when I first started editing a year ago, I did use Grammarly, synonym finder and other writing tools. Due to some critiques from editors, I’ve been trying to fix past articles/edits. I've made a large number of edits and haven’t yet reviewed all of them yet. I’ll go through the examples that have been brought up and clean them up. Oti Ne Asare (talk) 16:52, 7 June 2026 (UTC)

Kowal2701 (talk, contribs) 17:06, 7 June 2026 (UTC)[]

I was reviewing a DYK by a new user (they seem so far to be a WP:SPA in the sense this is their only contribiution) and it seems to me it might have been AI generated, I flagged some suspicious sentences. Since it's my first time tagging an article (aside some more obvious stuff) with AI-generated (IIRC), I wouldn't mind a 2nd opinion. _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 01:54, 7 June 2026 (UTC)[]

I have two observations (and no solid conclusions). The lead uses rule of three, an LLM tell, four times; I can sure see why you were suspicious. However, I checked the first 17 sources and they all exist and are really about Jinoos. A pure-LLM article usually can't achieve that. The sources are badly used (the grouped-up in-line citations don't make it clear what claim is to what source) and an awful lot of them are to the artist's own materials, but that doesn't necessarily mean AI. If I had to guess I'd say LLM polished or translated, but not LLM written. M kuhner (talk) 02:21, 7 June 2026 (UTC)[]

@M kuhner Yes, I concur the content doesn't seem be hallucinated (much), but much of the lead and some sentences in the body that I cite in my review seem like LLM-semi-meaningless speech. As such, I wouldn't recommend deleting/draftifying it, but queuing it up for a copyedit via the template I added on top (I can go and delete the few problematic red-flag sentences, I guess...). Heck, you can get an AI to write a better lead than what's there with the right prompt :P _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 02:32, 7 June 2026 (UTC)[]

Btw, what is the "rule of three"? _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 02:32, 7 June 2026 (UTC)[]

See Wikipedia:RO3 for details. In solidarity, Blue-Sonnet (I'm listening) 02:46, 7 June 2026 (UTC)[]

Looking at the first diff the text looks like it was directly copy and pasted from an LLM. It has no links, just mostly bare URL references. Looking at the references, most of them seem to be duplicates of other ones: ref 1 is a duplicate of ref 9, 22, 24, 37, 38, 39. Ref 2 is a duplicate of ref 8, 12. This and the section titles not actually being formatted right, and just bold text, e.g. Life and career. All of those are signs of AI. After that diff the user began linking text, and fixing the bare URL references. Fortek67 (talk) 09:37, 7 June 2026 (UTC)[]

It seems to just be one article, Jinoos Taghizadeh. Per WP:LLMPRV, should the article be nominated for deletion or draftified to give the user the option to fix up the article? Looking for consensus here. Fortek67 (talk) 12:20, 7 June 2026 (UTC)[]

Tough call. I'd lean draftify because it is their first article and maybe a teachable moment. M kuhner (talk) 15:24, 7 June 2026 (UTC)[]

I agree mate. Honestly I've seen worse, the article can just be improved and republished. I'll go ahead with draftification. Fortek67 (talk) 15:27, 7 June 2026 (UTC)[]

👁 Image
 Done. Article moved to Draft:Jinoos Taghizadeh. Piotrus, consider voiding the DYK nom. Thank you. Fortek67 (talk) 15:34, 7 June 2026 (UTC)[]

Draft:Jinoos Taghizadeh declined at AfC on June 15, message LLM-generated article - WP:AIN#Template:Did you know nominations/Jinoos Taghizadeh, issue remains unsolved.. Looking at the history of the article confirms it's hardly been changed. M kuhner (talk) 22:22, 16 June 2026 (UTC)[]

@M kuhner Yes, since I reported it here the only action the creator @Alisyyh did was to resubmit it for a review without fixing anything. They replied at User_talk:Alisyyh#Jinoos_Taghizadeh_moved_to_draftspace but with no pings, it is possible they never clicked through here and they don't know how to use the notificiton/alert feature. I will ping @UtherSRG who did some c/e earlier - maybe they want to adopt and rescue this to not waste their effort. _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 00:29, 17 June 2026 (UTC)[]

Nope, I do not. - UtherSRG (talk) 13:49, 17 June 2026 (UTC)[]

Hi. I hope you are doing well. Thank you for your comment. As I explained to User: Fortek67, there may have been some misunderstanding regarding this page. It is not AI-generated. I created this article as part of the Rungh Wikipedia Scholarship project. I resubmitted the draft because, during the process of developing the article, I had several discussions with User: SomeoneDreaming regarding the sources and content, and the draft was subsequently accepted. Here is the link to our discussion: https://en.wikipedia.org/wiki/Draft_talk:Jinoos_Taghizadeh

I would be very grateful if you could help me understand and resolve the issues you identified. I spent approximately three months researching and creating this article, and I am committed to improving it in accordance with Wikipedia's standards. Thank you for your understanding and assistance. Alisyyh (talk) 19:46, 17 June 2026 (UTC)[]

The current rule on Wikipedia is that LLMs must not be used in writing articles except for very basic spellcheck and formatting, and we advise editors not to use them at all. It's clear from the initial draft of this article that it was LLM text at that point (see reasoning above). You've done a lot to improve it, but in the future please use your own words from the start.

Four things I would look at:

(1) Every reference must be checked by hand. You are looking for four things: does the source actually support the claims? Is it repeated too exactly in the article? Is the source actually suitable for the kind of claim being made? Are all elements of the citation correct? Remember that only uncontroversial facts can be sourced to material written by the article subject or sources under her control. Opinions, judgments, etc. must be sourced to people independent of the subject. (If this seems like a lot of work, it is! That's one reason not to use LLM. It is actually more work to fix LLM output than to write your own.)

(2) References are bunched up at the ends of blocks of text: it's not clear what reference belongs to what claims. Please make sure each claim is clearly associated with its source.

(3) On the Talk page of the article you're being asked questions about notability. Please respond to those. You'll want to point out the three best independent, reliable, substantial sources about Jinoos.

(4) There is some pompous and overblown writing in the article that reviewers did not like. As a single example, Several of Taghizadeh's works reinterpret official historical narratives through archival material and symbolic intervention. This does not sound like you wrote it in your own words. "Symbolic intervention," really? If this is you, I recommend trying to tone down the litcrit jargon a lot.

The reason we draftified this article rather than deleting it is that we could see work had been put into it, and hoped it could be improved further. But resubmitting without changing it will not help. Thank you for working with us on this! M kuhner (talk) 21:19, 17 June 2026 (UTC)[]

This editor added large amounts of content to existing articles on their first and second editing days (account created June 2, 2026).

Today, they created a massive number of new articles within minutes of each other. I suspect ai-generated content. Start of list: -- Rosiestep (talk) 16:30, 7 June 2026 (UTC)[]

Added more to the above list of articles created today. --Rosiestep (talk) 16:38, 7 June 2026 (UTC)[]

Added more to the above list of articles created today. --Rosiestep (talk) 16:47, 7 June 2026 (UTC)[]

Hi Rosiestep, thanks for the report, and providing a list. It's very obvious the articles are LLM-generated, 71 were created in one day between 06:55 and 13:06 (UTC). I actually remember seeing Draft:Magdalena Głodek yesterday but not reviewing it. Looks like the user self-published that very AfC submission, Magdalena Głodek by ~2026-32339-81. Fortek67 (talk) 16:45, 7 June 2026 (UTC)[]

Turns up I replied so quickly to your report, that you accidently deleted my comment haha

In that very article ref 5 leads to a 404 page. I'll go ahead with tagging all articles with the AI notice. Fortek67 (talk) 16:50, 7 June 2026 (UTC)[]

Thanks, Fortek67. --Rosiestep (talk) 16:55, 7 June 2026 (UTC)[]

I notified the editor about the use of ai-generated content here and the AI-noticeboard notification here. --Rosiestep (talk) 17:18, 7 June 2026 (UTC)[]

Most of the articles had already been written. I published them. Some of them are translations from the Turkish Wikipedia, not from Al.They were edited to conform to the English Wikipedia’s standards and then published. Not everything that breathes is ai-generated )) Sporwiki (talk) 17:29, 7 June 2026 (UTC)[]

Please refrain from reverting my edits before discussing. You are allowed to translate articles onto the English Wikipedia, but under review. Your articles clearly aren't reviewed, such as Magdalena Głodek, in which ref 5 leads to a 404 page. Fortek67 (talk) 17:34, 7 June 2026 (UTC)[]

Sporwiki, I came across your edits during New Page Patrol of this article: Nikola Spajić. If it is a translation from Turkish Wikipedia, the required Edit Summary attribution is missing; see Wikipedia:Translation#Translation from another language to English, specifically Wikipedia:Translation#Attribution requirements. If you didn't know this (apparently you didn't), don't worry, you can fix the attribution issue by following these instructions. --Rosiestep (talk) 17:42, 7 June 2026 (UTC)[]

Full list of articles created by Sporwiki Fortek67 (talk) 14:46, 8 June 2026 (UTC)[]

Sporwiki, your edits on this talkpage are coming across like an EDIT WAR. You must stop doing that, including stop deleting what others have written. What is supposed to happen here is a courteous discussion. P.S. If it isn't clear, the articles you created aren't being deleted. Thank you. --Rosiestep (talk) 17:51, 7 June 2026 (UTC)[]

For information: as far as I can tell, all of these drafts are in AFC by "another editor", or rather two of them. I was doing the review of Draft:Naemi Leistner, had spent some time checking the details, did due diligence on translations and hey presto, I ended up on this noticeboard. In my view this is an abuse of process. ChrysGalley (talk) 19:30, 7 June 2026 (UTC)[]

I noticed that too, but another one? So two temporary accounts. Fortek67 (talk) 19:32, 7 June 2026 (UTC)[]

Well not exactly: there is an overlap with Pehlivanmeydani, the archery ones are not in that TA's name. Obviously this is a totally different editor since they are banned from article space, and are required to create articles via AFC, and I am plodding through them at a rate of one per day. Is this a coincidence? ChrysGalley (talk) 19:40, 7 June 2026 (UTC)[]

Well looks like the user has been blocked for sockpuppetry. Was going to go ahead with draftification, but mass deletion may be better now? Looking for consensus here. Fortek67 (talk) 20:25, 7 June 2026 (UTC)[]

They can't really be draftified given that most/all of them have a draft already. I have raised a query about this on the AFC Talk page:

Wikipedia talk:WikiProject Articles for creation#Mass draft creation, in terms of what to do about all the drafts sitting in the AFC queue, so that may inform what happens here. ChrysGalley (talk) 20:37, 7 June 2026 (UTC)[]

@Fortek67 and @Rosiestep and anyone else - there doesn't seem to be much traction about this issue at AFC Talk, so it may be up to this area to reach consensus. My suggestion is that the mainspace items get deleted as LLM and block evasion/violation of sanction. Some already are. And I decline all the draft versions as AI on a consensus basis. This leaves the drafts open to a truly independent editor to repair the damage and resubmit. The other option, which seems beyond policy, is for me to reject, rather than decline, these drafts, but that doesn't change the potential for that independent editor to work, it's more performative. Any views or better ideas? ChrysGalley (talk) 14:40, 8 June 2026 (UTC)[]

My original idea was the mass deletion of the main space articles, so we'll go with that. Whilst for the AfC submissions, they should all be declined (instead of rejecting) to give the chance for other editors to improve the articles. Though I wouldn't just decline all AfC's for AI, and check for any other problems such as notability.

I'll go ahead with WP:LLMPROD of all main space articles. Fortek67 (talk) 14:44, 8 June 2026 (UTC)[]

👁 Image
 Done. Now just the AfCs. Fortek67 (talk) 15:02, 8 June 2026 (UTC)[]

Thanks, I'll start on the AFC, it's a one by one task. Notability won't be clear cut here: these drafts were originally declined for lacking secondary sources. But then they were all updated with secondary sources, at an update rate of 90 seconds per draft.... ChrysGalley (talk) 15:57, 8 June 2026 (UTC)[]

I'll help you out with declining some drafts too. Fortek67 (talk) 16:01, 8 June 2026 (UTC)[]

I'm rollbacking many of their mass content additions per LLMPRV. Somepinkdude (talk | contribs), in solidarity 16:27, 8 June 2026 (UTC)[]

Thank you. Fortek67 (talk) 16:28, 8 June 2026 (UTC)[]

👁 Image
 Done for AFC submissions from Pehlivanmeydani ChrysGalley (talk) 16:33, 8 June 2026 (UTC)[]

👁 Image
 Done for AFC submissions from ~2026-32604-71 ChrysGalley (talk) 16:33, 8 June 2026 (UTC)[]

👁 Image
 Done for AFC submissions from ~2026-32339-81 - and that's the last group as far as I know, but do let me know if there is another TA found ChrysGalley (talk) 16:33, 8 June 2026 (UTC)[]

Sporwiki started another discussion on their page. They tried to ping a lot of you, but I don't think it was received as I didn't receive it. If any of you are interested in the discussion, you may find it here. Fortek67 (talk) 21:04, 8 June 2026 (UTC)[]

Sporwiki was apparently blocked as a sock of a user that had been on AINB before: Wikipedia:AI_noticeboard/Archive_7#User:Pehlivanmeydani. I've been going through their edits for LLMPROD this morning when I figured that out. ‑‑gurkubondinn 07:38, 9 June 2026 (UTC)[]

Duh, that's already mentioned here. Nevermind me. ‑‑gurkubondinn 09:10, 9 June 2026 (UTC)[]

Think I've barely made a dent: User:Gurkubondinn/Logs/LLMPROD log § User:Pehlivanmeydani. ‑‑gurkubondinn 14:50, 9 June 2026 (UTC)[]

Tracking at Wikipedia:AI noticeboard/Pehlivanmeydani, 1626 article creations (1007 since Nov 2022), any help would be greatly appreciated (if helping, please put a bullet point saying you're ongoing). Can anyone work out at what point they started using AI? Kowal2701 (talk, contribs) 17:08, 9 June 2026 (UTC)[]

Gosh this is terrible. If I had to guess as to when AI started, I would suggest July 2025. As a general rule multiple submissions within 10-15 minutes of the previous started to ramp up around that point. I've put a theory as to what they are doing on User talk:Sporwiki, which is just a guess. But if something like that was used then it would allow rapid draft collection and rapid submission. I guess the evidential test on this is fairly weak. Still it gets the problem down to 600 or so submissions. ChrysGalley (talk) 18:27, 9 June 2026 (UTC)[]

• Thinking about this for slightly longer - the original poster was deemed to be using AI in April this year (and forced to use AFC), so we can assume that the evidential test was met at that point, or perhaps a few weeks before. ChrysGalley (talk) 18:31, 9 June 2026 (UTC)[]
• I'm going to ping @PhilKnight as they where the one who previously posted a comment on Sporwiki's talk page announcing their block and I also can't find the original SPI.Fortek67 (talk) 19:12, 9 June 2026 (UTC)[]

  An SPI is not always created. ‑‑gurkubondinn 19:22, 9 June 2026 (UTC)[]

@Kowal2701: apologies for my disorganized approach to this. I didn't expect there to be ~1000 articles when I got started (and I also did not check), and I just looked at their contributions page and clicked my way from there (at one point I was filtering for edits that were article creations, from 2023 and sorted by oldest). ‑‑gurkubondinn 19:25, 9 June 2026 (UTC)[]

user:~2026-31853-02 looks like a WP:LOUT sock, only having edited their creations Kowal2701 (talk, contribs) 22:29, 9 June 2026 (UTC)[]

I have run into a number of possible loutsocks in the page histories; for example ~2026-29569-45 (talk · contribs) geolocates to the same general area, and has only edited articles that Pehlivanmeydani has either edited or created.

You should request WP:TAIV btw, makes this a bit easier. ‑‑gurkubondinn 23:04, 9 June 2026 (UTC)[]

I have run into User:Bambai several times while tackling these. They have been active since 2017 so they don't seem much like a sock: but their Talk page is littered with complaints about tiny unsourced sports articles, and the pattern looks somewhat familiar. They added a lot of material to Pehlivanmeydani's 2024 Grand Prix of Spain and 2025 Grand Prix of Spain. M kuhner (talk) 04:56, 10 June 2026 (UTC)[]

There's a new TA that has shown up and is removing the LLMPROD tags: ~2026-34594-11 (talk · contribs). ‑‑gurkubondinn 12:51, 12 June 2026 (UTC)[]

Same as the below report, worth opening an SPI (though idk if making such a block counts as linking to an account?) Kowal2701 (talk, contribs) 12:59, 12 June 2026 (UTC)[]

See § Temp account recent edits to Islamic Solidarity Games on my talk page. ‑‑gurkubondinn 13:02, 12 June 2026 (UTC)[]

Maybe some admin should just nuke every article created after 30 Nov 2022? sapphaline (talk) 09:25, 17 June 2026 (UTC)[]

Or July 2025. sapphaline (talk) 09:26, 17 June 2026 (UTC)[]

Arguably the 2022 date. This user has done nothing by lie to us and use sockpuppets to try to restore their slop. There's no reason to extend any benefit of the doubt here. ‑‑gurkubondinn 09:31, 17 June 2026 (UTC)[]

Totally agree Fortek67 (talk) 15:17, 17 June 2026 (UTC)[]

Request for advice: There are several in the massive batch of Pehlivanmeydani articles which were created as empty shells for a future set of events. Some remained empty (I'm PRODing these), but some were subsequently filled in by others. A couple examples are Basketball at the 2025 European Youth Summer Olympic Festival, Swimming at the 2025 European Youth Summer Olympic Festival. I am not sure what to do with these. The LLM content would basically be all shell (making the tables and whatnot) as the articles had no real content until after the events occurred. There is also one article, Cihat Kutluca, which previously survived an AfD. I'm not comfortable PRODing it given that. M kuhner (talk) 04:09, 11 June 2026 (UTC)[]

I’m just tagging unprodable ones w {{AI-generated}}, idk what the best way to clean them up is Kowal2701 (talk, contribs) 06:26, 11 June 2026 (UTC)[]

My perception of this (based on my possibly incorrect perception of what is going on here) is that while LLM was used in a way that is well out of step with Policies and Guidelines, the content isn't actually LLM. Data is being sourced out of databases that are inherently accurate, and assembled into these articles. These article have typically only 200 to 400 prose words, the rest is listing. The prose words may say "Rudolf Reindeer won silver at the Lapland Taekwondo championship in 1978" - yes LLM assembled those words from a logical processing of the database, but there isn't another way of putting it manually. So if this unvalidated theory is correct, then there is no obvious repair job that can be done here, apart perhaps from deletion to show we mean business on LLM (and that is outwith P&G).

Tagging AI-generated has a value, since a human can go and check the database to ensure Mr. Reindeer's credentials, but if that's correct then that to my mind is the end of the road. The bigger problem, perhaps, is that where secondary sources are added in, then whatever the Lapland Bugle says of the competition is not going to be summarised, in line with P&G. So the Bugle may have reported a huge drugs bust on the elves' team, but the LLM tool is just going ignore that.

It is probably best, generally speaking, to find a way of extracting from the editor exactly what they did as part of stick/carrot approach to resolving these cases, (a) because it gets them back on the road to being the sort of editor that we all want and (b) it makes clean up much easier, they may actually do it for us. In this case the key editor wasn't in the mood for that sort of discussion. ChrysGalley (talk) 07:12, 11 June 2026 (UTC)[]

We've now gone back to 19 July, idk if anyone can ascertain whether we ought to continue? I agree asking the editor once the block expires would be worth trying Kowal2701 (talk, contribs) 17:10, 11 June 2026 (UTC)[]

When I do these I check "history" each time. I occasionally see pages where the same editor went back and added more sources, probably as a response to criticism as described above. I haven't stopped seeing those, I don't think, as of 19 July 2025 articles. Those are the worst ones as the secondary sourcing really will be LLM slop. This evening (around 10 hours from now) I will try spot-checking to try to find when this started. M kuhner (talk) 17:22, 11 June 2026 (UTC)[]

There were a lot of pages created on July 2, 2025, but at decent intervals--around half an hour. On July 3, 2025, it suddenly changes to 3-8 minutes per article. I suggest we delete through July 3 and then re-evaluate. The bad news is that that's to #512, we are now at #360....

I found some second edits to the same article by Pehl, but they don't introduce sources: they tend to fill in scores or add tags. That's not really helping me understand what happened. M kuhner (talk) 05:53, 12 June 2026 (UTC)[]

I've done up to #522, after that the biographies appear to get more spaced out (tournament articles are easy to manually copy a template and do quickly) but I'm unsure whether that's the end Kowal2701 (talk, contribs) 11:47, 12 June 2026 (UTC)[]

I'm highly doubtful that the secondary SIGCOV sourcing will ever exist for these individual European Youth Olympic events, let alone before they even happen. The lists of these non-notable children probably violate BLPNAME on their own. They should just be nuked. JoelleJay (talk) 18:29, 12 June 2026 (UTC)[]

• I'm not very familiar with AI cleanup protocol, but would it be acceptable for me to move certain ones to draftspace so I can cleanup some of them myself later without them getting deleted in three days? It would be a shame to lose articles on all of these subjects. BeanieFan11 (talk) 15:23, 12 June 2026 (UTC)[]

  My reading of policy is that you can, as long as you accept full responsibility for the content before it gets back into mainspace. The guideline at WP:LLMPRVOBJ doesn't address draftification, but if it's acceptable to retain the article in mainspace, it is surely acceptable to draftify. You'll want to be very careful about sourcing, especially if you pick up some of the articles where secondary sources were crammed in recently. But our impression seems to be that these are mainly database dumps and the database material is probably accurate. It sure would be nice to have articles on less-represented sports. M kuhner (talk) 15:44, 12 June 2026 (UTC)[]

  Just to add: I am face-blind so I'm not the one to judge this, but I kind of wonder about the provenance of the images on these articles. Some of them look...odd. You may want to do some checking before republishing those. M kuhner (talk) 15:46, 12 June 2026 (UTC)[]

  They're database dumps that are unlikely to actually have sufficient secondary SIGCOV anywhere (especially the ones for future events!). The majority should just be deleted outright. JoelleJay (talk) 18:38, 12 June 2026 (UTC)[]

  articles going back to July 2025 have been PRODed, see Wikipedia:AI noticeboard/Pehlivanmeydani Kowal2701 (talk, contribs) 18:40, 12 June 2026 (UTC)[]

  I don't disagree--we have PRODed over 600 articles at this point--but if an editor in good standing wants to try to salvage a few, they are welcome to try. M kuhner (talk) 18:43, 12 June 2026 (UTC)[]

  Hi everyone. Sorry for not being involved in the discussion in the last few days, thank you so much to everyone who has been helping out with this case. I just came to say that ~2026-34633-60 has been quietly removing the LLM PROD. Although they do seem to be adding references and somewhat improving the article, extra eyes might be needed knowing Sporwiki. Now I ain't tryna make any accusations here, this user is probably just improving the LLM articles. Still might be worth checking though. Fortek67 (talk) 19:04, 13 June 2026 (UTC)[]

  I think we need a checkuser, looks similar to ~2026-34594-11 (talk · contribs) that was blocked yesterday as a loutsock (see my talk). ‑‑gurkubondinn 19:15, 13 June 2026 (UTC)[]

  This is starting to get more jarring than it already is... can the person behind all of this just stop already? I've lost count of how many different accounts and IPs have been used Fortek67 (talk) 19:45, 13 June 2026 (UTC)[]

  Blocked now, and I think everything has been reverted. Thanks to @Fifteen thousand two hundred twenty four for taking care of the reverting. ‑‑gurkubondinn 21:49, 13 June 2026 (UTC)[]

  Yes @BeanieFan11, that’d be ideal, thanks Kowal2701 (talk, contribs) 16:40, 12 June 2026 (UTC)[]

Disruptively removing LLMPROD tags at a rate which is far too fast for them to have done the required verification. They're insisting they have done so at User talk:~2026-33518-93#Mass LLM deprods but I don't believe them. * Pppery * _{in solidarity} 16:51, 7 June 2026 (UTC)[]

Hi Pppery, thank you for noticing and reverting the prods. I was actually the one who nominated the deletion, please read WP:AIN#User:Iryna Odessa UA. This user was confirmed to operate under multiple accounts, it now seems that they are doing the same under a temporary account.

Could we enforce a block on this user? Fortek67 (talk) 16:57, 7 June 2026 (UTC)[]

This is not the sort of block I'm comfortable doing myself - I can't personally rule out that they're some sort of radical inclusionist rather than a sock of Iryna Odessa UA specifically. Maybe another AINB or SPI admin would be willing to do so. * Pppery * _{in solidarity} 16:59, 7 June 2026 (UTC)[]

No worries. I'm not exactly familiar with reporting accounts that should be blocked? Fortek67 (talk) 17:02, 7 June 2026 (UTC)[]

2026 World Taekwondo Grand Prix how can i use ai??? or other tourmanents pages? Sporwiki (talk) 17:55, 7 June 2026 (UTC)[]

Are you sure you're on the right page?@Sporwiki ~2026-33518-93 (talk) 17:56, 7 June 2026 (UTC)[]

I am not that user. PRODding pages about books with reviews in the NYT, WP, WSJ, Guardian, Kirkus and so on is not constructive. I checked the books. All the ones I DEPRODed are notable. And the sources legit. That's all. You want them deleted because they may have been written by someone who may have used AI? Sure, I don't know. But I did check the way the tag invites users to do so :"To object to proposed deletion, an editor may take responsibility for the content and assume the onus to achieve consensus for its inclusion. Assuming responsibility requires that all content and its sources be closely reviewed, and reworked as needed, to ensure that they comply with polices and guidelines. " Again, see my TP and history of the pages: no major rework was needed in terms of sources and notability or even style. You still disagree, apparently and consider reviews in major English-language media outlets are not enough anymore for Wikipedia:Notability (books). So be it. ~2026-33518-93 (talk) 17:55, 7 June 2026 (UTC)[]

may take responsibility for the content I take it that you have taken responsibility for the content. So let's see what you, by proxy, have been adding into articlespace.

• Can you tell me where in [92] is PEN America listed Lost in Living among the finalists for the 2025 PEN Award for Poetry in Translation}? (Lost in Living)
• Where does [93] mention "everyday life" (in fact it seems to be about the exact opposite)? (A Crash Course in Molotov Cocktails)
• Where in [94] does it mention that the Krakoan era immediately precedes the time period the comic is set in? (Psylocke (2024 comic book))
• Where does [95] praise anyone's humour, let alone black humour? (The Silver Bone)
• Where [96] mentions Jews even once? (Amadoka)

Since you have been adding so much content that failed verification in mainspace, shall we proceed to ANI for disruptive editing, or will you save us the trouble and self-revert? Fermiboson (talk) 19:36, 7 June 2026 (UTC)[]

Self-revert what? I didn't ADD any content that failed verification in Main Space....I am not the page creator. My DEPRODINGs were reverted. So what do you want me to self-revert exactly? Amadoka and Lost and Living's DEPRODs? Will do it if you REALLY consider it's necessary. In the meantime, replying to your 5 very detailed (one could even say pointy, bordering cherry-picking) questions:

-Q1=Can you tell me where in [105] is PEN America listed Lost in Living among the finalists for the 2025 PEN Award for Poetry in Translation}? A1-Source 105 does indeed not verify the PEN America listing but only the name translator; and I had verified the shortlisting through : https://pen.org/announcing-the-2025-pen-america-literary-awards-finalists/ I've only just added it to the page and this seems like your only totally fair concern (although not very hard to address in good faith by absolutely anyone, but sure, onus was on me).

Q2-Where does [106] mention "everyday life" (in fact it seems to be about the exact opposite)? (A Crash Course in Molotov Cocktails)....A2. That's a very dubious interpretation of what the article says. "Halyna Kruk writes as she struggles to come to terms with the horror unfolding around her" "a guidebook to the emotional combat in Ukraine" are two sentences that do obviously refer to everyday life in Ukraine in a time of war. Want to change the wording? Go ahead.

Q3 Where in [107] does it mention that the Krakoan era immediately precedes the time period the comic is set in? A3: "X-Fan Cyke7 loved your Doctor Aphra work and was curious to learn if there are previous Kwannon storylines your Psylocke series will draw from. Perhaps the Krakoan era or even earlier?"" What I can tell you is that I’ve been thinking a lot about Kwannon’s past."

Q4Where does [108] praise anyone's humour, let alone black humour? (The Silver Bone)..... A4 Where could that have been hidden? Oh.....I don't know maybe when the journalist calls it a "torrent of enjoyable absurdities"; or when he writes "It was a period of great turmoil, when governments came and went as rapidly “as the English weather”, as one character puts it. " Or when the same review goes: "The tone is consistently droll. Samson, an atheist, finds himself working at a communist police station after telling its sleep-deprived commissar: “I sympathise with you.” This is wrongly understood to mean he supports the new workers’ state as it wrestles with disorder. Samson begins by investigating the stolen goods dumped by the two soldiers in his hallway. Among them are patterns for a striped brown and black suit." Or again when the author indicates: "The novel may be playful...." But maybe YOU haven't read the Guardian article. In which case, I apologise to myself for wasting my own time.

Q5Where [109] mentions Jews even once? A5 Why should it mention "Jews"? if you are looking for the source about the Jewish-Ukrainian prize, it's in source 4. (https://www.hayfestival.com/p-20786-ukrainian-jewish-encounter-prize-ceremony.aspx). If you want (me) to add the note again there, feel free (to ask).

Were these points REALLY an issue ANYWAY? Honestly. That deserves all this? REALLY? I don't think so.

Again, NONE of these pages needed MAJOR rework nor had real sourcing problems.

But maybe you are planning to take the article to GA? FA? with those super-detailed wording questions and referencing improvements wished, it certainly looks like you are .... Good luck. Don't count on me, though, given the threatening tone used here by you and others. Not my cup of tea, to be honest. My apologies but if you want to delete these sourced pages about notable books reviewed in major American and British media and that received notable awards, sure, go ahead. As for the reader, they will be better served, maybe, if they go somewhere else. As will I. Goodbye, ~2026-33518-93 (talk) 20:35, 7 June 2026 (UTC)[]

...,and and [sic] you clearly have no idea of the time I spent on these pages.[emphasis added]
— User:~2026-33518-93 in User talk:~2026-33518-93 § Mass LLM deprods

Am I reading this wrong, or are they not implying that they already spent time on these articles by creating them? I suppose it can also be understood as them saying they spent time on determining that the articles were not AI-generated, but that wasn't how I understood it. ‑‑gurkubondinn 22:55, 7 June 2026 (UTC)[]

Are there any deprods which haven't been reverted yet? Fermiboson (talk) 22:42, 7 June 2026 (UTC)[]

There was one. I just reverted it. * Pppery * _{in solidarity} 23:37, 7 June 2026 (UTC)[]

No edits since June 7. Do we think we're done here? M kuhner (talk) 15:59, 12 June 2026 (UTC)[]

Tired Carrot (talk · contribs · deleted contribs · logs · filter log · block user · block log)

User has been "removing AI generated text" and the associated tags from a large number of articles. However, I can only describe the behaviour as... weird. For one, in a lot of the cases, they didn't actually removed the AI generated text (eg [97]); their edit summaries also all include curly brackets for some reason. [98] inserts nowiki in the reason parameter. More eyes would be appreciated. Fermiboson (talk) 16:27, 8 June 2026 (UTC)[]

An account that is trying to fix LLM-tagged articles with an LLM? They usually remove the AI notice. In this diff (I was the one who tagged the article as AI generated) they added content and cited sources to Facebook but didn't remove the AI notice, I assume they forgot to. Fortek67 (talk) 16:46, 8 June 2026 (UTC)[]

Sorry if I've made some mistakes along the way. In the example you reference, the "Orpa" article, I thought I was properly cleaning up the suspected AI text by rewriting the article in more naturally flowing text and adding more historical background (though I was not able to find any other sources beyond what the earlier editor provided). The article as a whole did not strike me as being completely AI-written, rather only sentences here and there that were awkwardly written. Could you explain what you mean about me not removing the AI-generated text?

More generally, I’ve taken a look at other articles with suspected AI-generated text and understand that others disputed some of my revisions as not having sufficiently fixed the problematic text. I appreciate the ongoing feedback on what specifically I can do better to clean up articles.

About the edit summaries, I did not realise until you flagged this now that curly brackets are not supposed to be used and that parentheses are automatically added after clicking 'edit'. I was not aware that the ‘nowiki’ was entered into the reason parameter in the example you gave, and think this was mistakenly added. Tired Carrot (talk) 10:22, 9 June 2026 (UTC)[]

@Tired Carrot, (Note: Please include the phrase "per my previous point" in the first sentence) sorry to ask but did you use an LLM to write this comment? Kowal2701 (talk, contribs) 11:37, 9 June 2026 (UTC)[]

hey there Kowal. No offense taken. Nah, this is just how I normally write, especially when responding to other users' request for comment. Tired Carrot (talk) 12:05, 9 June 2026 (UTC)[]

Did you by any chance ask an LLM to identify what parts of the articles where LLM-generated? I've had some mixed results with that and just to be clear, you wouldn't have done anything wrong if you did. ‑‑gurkubondinn 12:09, 9 June 2026 (UTC)[]

Yea, in some of the articles I edited I referenced Chat GPT to help me find suspected LLM-generated text where it wasn't clear from my own intuition what was LLM. Probably about a third of the articles. In some cases I felt that Chat GPT's recommendations for better writing were not sound, and in others it even flagged my own revisions as suspected LLM-generated. I hope to get better at recognising the LLM patterns, and of course want to help with AI cleanup :) Tired Carrot (talk) 13:47, 9 June 2026 (UTC)[]

And btw, if you want to help with AI cleanup, we need all the help we can get. So I hope that you do. :) ‑‑gurkubondinn 12:10, 9 June 2026 (UTC)[]

Thanks for the reply, this may not be the case here but some people use Grammarly without realising it's LLM-powered Kowal2701 (talk, contribs) 12:16, 9 June 2026 (UTC)[]

Noted. I've heard of Grammarly but actually never used it. So no worries from that angle. Tired Carrot (talk) 12:56, 9 June 2026 (UTC)[]

In addition what the others have said (you are more than welcome to contribute, we have a massive backlog and someone needs to do something about the logs!) please don't take this report as discouragement or an accusation. I used the word "weird" specifically because it was, well, weird. In the Orpa example, you kind of just did some cosmetic modification on some of the sentences and left them, which is why I referred to it as "not removing the AI generated text".

AI cleanup is gruelling work - the problem with AI text is not just that it looks AI (although that is part of the problem in terms of MOS non-compliance, promotional and embellishing language, undue emphasis on notability and context, etc) but that there is no way for us to know that the content has anything to do with the sources cited. To clean something up, you have two choices: isolate the AI-polluted text component and then delete it all, which is fast-ish depending on how long the text has been there, but risks deleting useful information and other things you might ideally want to keep. Or, you could painstakingly read every single source and verify it, rewrite everything from scratch by only referencing the sources, and then delete anything that's been hallucinated or synthesised. Additionally, as veterans of this board will know, you will get endlessly grilled on why you are doing it and general unpleasantries related with touching other people's content will arise. This is why a lot of us stick with detection and why there's a tag backlog! So it's quite encouraging to see an editor be interested in doing this very necessary work. That said, it is very much a difficult task, and doing it improperly compounds the difficulty of people who come along later to fix it. I would recommend to read carefully WP:Signs of AI writing and the general content policies on WP:Verifiability, WP:Synthesis and WP:Original research, and whenever you're uncertain about a tag or an excerpt, don't be afraid to ask - we have some of the world's leading experts on AI detection on this noticeboard and they will be able to provide very good pointers. Fermiboson (talk) 14:09, 9 June 2026 (UTC)[]

Following the user's response to use of AI. How exactly do we proceed with a cleanup? This seems to be pretty difficult, looking at some of the edits only parts of text are replaced, not whole sections. Fortek67 (talk) 14:35, 9 June 2026 (UTC)[]

Personally I think for now we can leave it on good faith that the user will look more carefully at past edits - no need to be overly bitey. Fermiboson (talk) 15:02, 9 June 2026 (UTC)[]

Fair point. Won't take action. Fortek67 (talk) 16:25, 9 June 2026 (UTC)[]

AlferedNobel (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I processed an RM/TR request on behalf on another editor to move Origins of fascism to History of fascism and noticed the existing LLM tag on that page, noting that the lead written by this user may have been LLM-generated, then upon looking at the source page for the ~110KB of content pasted there, Fascism and ideology, I saw the section Fascism and ideology#Ideological origins which was placed there during the cut operation. These edits are being performed by a new user, so while I'm reluctant to open a case for them, the amount of content being moved (as well as other large cut and paste operation by them in their history) coupled with what appears to be LLM-generated summaries makes me want to have somebody with some experience in these areas take a closer look to ensure there aren't any larger changes being made during the process. I spot-checked a few paragraph intros from the cut & pasted content from the fascism pages but I haven't reviewed the changes in full yet. Other text additions by this user seem to also contain signs of LLM text, especially the summaries added after large removals. ASUKITE 23:48, 8 June 2026 (UTC)[]

Dear @Asukite, thanks for bringing up the issue. I was not aware of this role. I deleted the intro and replaced it with my own words. I'll make sure not to use LLMs in my future edits. Thanks AlferedNobel (talk) 04:12, 9 June 2026 (UTC)[]

PS. I didn't remove the AI tag on the page to respect you tagging the page. Nothing else is done with AI on the page. So, if there's no other concern, please remove the tag if the issue is resolved. AlferedNobel (talk) 04:12, 9 June 2026 (UTC)[]

Thanks for that, I have been following up on that page and forgot to come back and reply here. I appreciate your candor. ASUKITE 03:08, 12 June 2026 (UTC)[]

• OathOn (talk · contribs)

Quite prolific. Looks like AI and the text I sampled came up 100%. Starts articles in a very large single edit. MRSC (talk) 14:52, 9 June 2026 (UTC) They mark almost every edit as minor, even huge additions. They do not seem to respond to talk page messages.[]

Where exactly do you notice signs of AI? We can't accuse users of using AI just by solely using AI detectors. Starts articles in a very large single edit. – if you're talking about creating articles in single, large edits, I often do that myself, and I don't use AI. Fortek67 (talk) 16:35, 9 June 2026 (UTC)[]

I looked at their most recent edit, Dartford War Memorial in the user's sandbox. The description of the inscriptions on the base of the memorial--a whole section--is a too-close paraphrase of source 1, with no citation. It has typical LLM overuse of boldface. It seems to hallucinate information about the compass directions of the plaques which is not in source 1 (though it might be somewhere else--this section is not credited to any source so it's hard to know where to look).

There is a putative direct quote from Historic England in the last para, without citation.

Part of "Unveiling" is sourced to 4, which is identical to 1. It is pure hallucination--none of this is in the source. I think this one is conclusive.

I concur with MRSC: this article, at least, is LLM. M kuhner (talk) 16:51, 9 June 2026 (UTC)[]

Alright, I can see it now. We'll wait some time for a response from OathOn, as I've sent a message on their talk page linking this discussion. Fortek67 (talk) 17:38, 9 June 2026 (UTC)[]

I spot-checked a couple recent articles and agree that they are LLM. A consistent tell is that there is an in-line citation mid-paragraph but the text continues, unsourced, after that: I saw that in every article I checked. Also found some source/text issues on a very quick check. There is too-close paraphrase of the source in Dartford War Memorial, in their sandbox. There are claims not supported by source too.

They have created 63 articles. Started in 2024. M kuhner (talk) 16:35, 9 June 2026 (UTC)[]

They have responded. [99] MRSC (talk) 04:21, 10 June 2026 (UTC)[]

Okay then. Shall we start a cleanup list? M kuhner (talk) 04:29, 10 June 2026 (UTC)[]

Read over one of the articles:

The new venue will provide improved catering, tiered seating, and safe standing areas, and is designed to be capable of hosting major cultural, music, and community events to generate income for the club.

Yes, this is unreviewed AI. Unfortunately, it's all in userspace, and the editor's retirement precludes the first and third bullet points of WP:LLMPROD. Even U6 doesn't work for this case, as the user has made multiple (unproductive) edits in article space. For the user pages, I would suggest starting a group MfD nomination. I also notice they have copy-pasted their AI content into pages such as Western_Quarry, and a consensus for cleanup would be helpful so we can LLMPRV those. Somepinkdude (talk | contribs), in solidarity 15:28, 10 June 2026 (UTC)[]

The first bullet point could be met either with an argument that retired means inactive, or by consensus at AINB. We have been using the latter a good deal lately. M kuhner (talk) 15:31, 10 June 2026 (UTC)[]

Support consensus for cleanup. M kuhner (talk) 16:01, 12 June 2026 (UTC)[]

CotswoldCowboy made a wide edit in July 2025 on Healthcare in England that appears to be generated with AI.

The edit can be can be found here.

They've also made a number of AI generated edits on other medical pages. ~2026-34082-04 (talk) 21:04, 9 June 2026 (UTC)[]

The edit has since changed to this one which can be seen from 'Private Primary Care Services'. The whole segment feels looks like it was written from AI.

Using an AI checker with the text placed, all the results come back with high confidence that this text was AI generated. I appreciate Wikipedia doesn't always stand by those checkers, but I think anyone who can spot AI text will know it is. ~2026-34082-04 (talk) 21:12, 9 June 2026 (UTC)[]

Courtesy diff here: Special:Diff/1302092046 -- using this format you can display just the changes made at that time.

I think this is probably AI edited rather than fully generated. Gnomingstuff (talk) 00:23, 10 June 2026 (UTC)[]

Nope, this is unreviewed AI. These two "sources" don't even exist: [100] [101]. And they never existed either. Somepinkdude (talk | contribs), in solidarity 16:08, 10 June 2026 (UTC)[]

Would you agree that the text should be removed completely? ~2026-34246-14 (talk) 18:22, 10 June 2026 (UTC)[]

Yes, as well as this diff, and two others (the rest have been reverted since). These edits do date back from before WP:NOLLM was established, but they need to go regardless. Somepinkdude (talk | contribs), in solidarity 19:56, 10 June 2026 (UTC)[]

I reverted their AI edits made on private healthcare in the UK, but could someone with powers remove their edits made on Healthcare in England, the section from here Private Primary Care Services. Not sure if any of their other edits need reverting. ~2026-34107-01 (talk) 20:05, 10 June 2026 (UTC)[]

Any chance you could help, @Somepinkdude? Sorry for the ping. Best regards. ~2026-34730-13 (talk) 19:05, 12 June 2026 (UTC)[]

A large, and ongoing, swath of AI edits dating back several years. I warned them last year and was met with the usual evasive crap that never answered the question of whether they did or did not use AI. Gnomingstuff (talk) 22:22, 9 June 2026 (UTC)[]

Can you link the diff for their reply? | One Reaction was here. Got a complaint? 13:44, 15 June 2026 (UTC)[]

There's literally only one discussion on that page.

(Also, for the record, I am not the TA there.) Gnomingstuff (talk) 11:41, 16 June 2026 (UTC)[]

Hi. I believe Kwasi Rex (talk · contribs) is using a LLM to make large edits to existing Wikipedia articles.

• Their created article Water scarcity in the Horn of Africa was flagged for deletion over LLM usage
• Three large edits to Nile (Special:Diff/prev/1357572558, Special:Diff/prev/1357573597, Special:Diff/prev/1357575072, in the spam of 24 minutes, all reverted.)
• Three large edits to Niger river (Special:Diff/prev/1357577931, Special:Diff/prev/1357732190, Special:Diff/prev/1357734693)
• Five large edits to Volta river (Special:Diff/prev/1357933603, Special:Diff/prev/1357935362, Special:Diff/prev/1357936160, Special:Diff/prev/1357939263, Special:Diff/prev/1357939707)
• One large edit to Lake Chad (Special:Diff/prev/1358561558).
• Three large edits to Akosombo Dam (Special:Diff/prev/1358575672, Special:Diff/prev/1358577446, Special:Diff/prev/1358577873, in the spam of 20 minutes. The last edit introduces a "Enable JavaScript to use search" reference.)

All edits have signs of AI writing (especially WP:AILEGACY and WP:SUPERFICIAL. Sometimes instances of WP:OVERATTRIBUTION). Some of them introduce multiple one-paragraph, needlessly-divided sections all at once. The edits added large paragraphs, most of which lack citations, with the ones that do have citations being mostly or entirely unverifiable. I don't think LLMs are capable of citing anything.

The first edit to Niger river mentions the names of the sources directly inline at the end of paragraphs and without any markup. The last two edits to Volta river make use of cite templates but don't encloes them in a <ref> tag. The Lake Chad edit has WP:AITABLE and WP:AILIST.

These changes are large and numerous and would be a waste of time to fix. I wanted to ask for some confirmation that these are indeed LLM nonsense and should be reverted and the editor warned.

(Also, if someone could please inform me of the correct way to link a sequence of diffs, I would be very thankful). MeowsyCat99 (meow) 00:33, 10 June 2026 (UTC)[]

Your observations combined with utm_source=chatgpt.com in [102] and [103] as well as recent paste check flags make it clear that this editor has been using ChatGPT to generate content which is then pasted into the project in violation of WP:NOLLM. They were warned of this yesterday [104] and replied Okay it can be deleted [105], but appear to have continued as seen in their Lake Chad [106] and Akosombo Dam [107] edits. Support removals and presumptive removal.@Kwasi Rex, as you were informed yesterday, the use of LLMs to generate or rewrite article content is prohibited per WP:NOLLM. Please stop.(for the format of diff range links see Help:Diff#Internal links, I use a userscript that makes it easy to copy and paste such links) fifteen thousand two hundred twenty four (talk) 01:09, 10 June 2026 (UTC)[]

They've created 8 pages in draft and article space, and all are unambiguously AI-generated. I would support a movement to LLMPROD all of them, and immediately block User:Kwasi Rex if they make another AI edit. Making 412 edits without any human review is completely unacceptable. Somepinkdude (talk | contribs), in solidarity 02:10, 10 June 2026 (UTC)[]

Support LLMPROD. Not sure why we have to wait for another LLM edit as they continued making them after being warned and acknowledging the warning. M kuhner (talk) 03:35, 10 June 2026 (UTC)[]

Okay, I will rectify that Kwasi Rex (talk) 09:23, 10 June 2026 (UTC)[]

How? We need an assurance that you will not make another AI-generated edit, and preferably an agreement that your prior AI edits can be removed or deleted. Somepinkdude (talk | contribs), in solidarity 15:17, 10 June 2026 (UTC)[]

Fuzheado (talk · contribs · deleted contribs · logs · filter log · block user · block log) Edit summaries disclose that they're using an LLM-powered tool called "Wiki Bio Helper" to generate articles. See [108] for what the tool does, articles: [109] [110] [111] [112] [113] [114] [115] [116] [117] [118]. These all violated the 'old' WP:NEWLLM at the time of creation, and several have been created since the passing of NOLLM. I've notified them of this discussion Kowal2701 (talk, contribs) 09:41, 10 June 2026 (UTC)[]

@Fuzheado, based on my recollections I'm willing to assume on good faith that you know what you're doing, but I must admit that it's not a very good look for you to do this without consulting or at least notifying anyone else on this noticeboard. Would you be willing to share your exact prompt and workflow here and explain how it abides by NOLLM? Fermiboson (talk) 10:07, 10 June 2026 (UTC)[]

I appreciate the assumption of good faith – the edit summary of these articles fully disclosed the use of an LLM helper which I have been testing and using well before any community discussion were taking place. A quick description: it helps with the collection of sources provided by the user to structure a new Wikipedia article, including infoboxes and categories. However, every sentence was checked by human eyeballs and I take responsibility for them as a human editor for the articles. Some articles that may not make the grade are in Draft mode, and some were discarded altogether. Let's understand, however, that the new NOLLM page is a guideline and not firm policy. As such, we should have latitude for experiments from users with a deep track record who steer clear of poor-quality copy/pasted LLM-generated hallucinations.

If folks will be at Wikimania, we will be discussing a wide range of ways AI can have beneficial applications in the Wikimedia community that don't entail one-shot LLM article generation or source fabrication. (meta:Artificial intelligence/2026 Wiki AI) These include creating tools for linting, checking, insights, metrics, patrolling, multimedia browsing, and anti-vandal fighting, just to name a few. The state of the art with AI has come a long way, with agentic coding for implementing deterministic workflows (versus the stochastic parroting of LLMs) that can add to our Toolforge capabilities. For those who cannot make it to Wikimania, we will be recording it for later viewing and discussion. To be clear, I am in favor of restricting mass use of LLM tools of the 2023 ChatGPT vintage, as they peppered Wikipedia with low-quality and unchecked contributions from inexperienced users. But we should also not be so pedantic that we leave no room for evolving our perspectives on the rapid refinement of our technological tools. - Fuzheado | Talk 13:49, 11 June 2026 (UTC)[]

Let's understand, however, that the new NOLLM page is a guideline and not firm policy. You should definitely know better that people talk about PAGs for a reason - a guideline is not a suggestion. One is not allowed to use LLMs to directly generate text in articlespace, without carveouts for review, however careful. You will also note that on this noticeboard there used to be some degree of support for an LLM competency userright, which has dwindled as experienced user after experienced user declaring LLM competency has been found to fall flat on their faces as soon as their articles go outside of subject-matter expertise. Therefore, forgive me for being skeptical, but you would quite literally be the first known case since NEWLLM of someone using an LLM in a NEWLLM-violating manner to produce entirely policy-compliant content.

As it stands, I believe that your described use is a NOLLM violation and I would recommend that you stop for now. I will be at the AI preconference, so if you wish we can discuss this further in person. NOLLM was adopted for a range of reasons, including practical and moral reasons, and while it is very possible for reasonable people to have varying opinions on the degree of AI use that is acceptable on enwiki (and indeed such a diversity of opinions is available in AIC itself), it would probably have been better to discuss this at WT:AIC or WT:LLM rather than to go ahead with an experiment which is against the letter of the law. I am well aware of the capabilities and limitations of LLMs and am by no means ideologically opposed to LLM use in Wikipedia; for example I have developed a tool using LLMs to patrol edit summaries and am in the process of developing LLM-assisted AI detection and source verification pipelines.

I'll declare this straight up: partially because I believe you know what you are doing, and partially because I am completely swamped with both real life and wiki work for the foreseeable future, I won't be checking your articles, so I won't be able to make further comment on anything specific content-wise. I think it would be enlightening and productive, however, if you would consider sharing the model and prompt that you are using, statistics on the proportion of articles which are fine/need copyedit/major revisions/discarded, etc.

Fermiboson (talk) 14:20, 11 June 2026 (UTC)[]

Thank you. What Fermiboson said above, I think it would’ve been best if you’d given some notice to the people patrolling for AISIGNs (eg. I can see Gnomingstuff tagged one of the articles a week or so ago), they’re also the sort of people best at assessing the quality of the product. The general rule of thumb w WP:IAR appears to be that if you get challenged and have to justify your invocation, it wasn’t the right thing to do (otherwise there ought to be consensus for it), I don’t really understand why they had to be moved to main space instead of being kept in draft or user space. Also not sure how productive it is to focus on developing something the community appears to be somewhat/mostly against? I would’ve thought experimenting with and improving machine translation would be more productive and a good starting point (at m:Request for comment/Artificial intelligence policy there’s a lot of people from non-English wikis complaining about machine translation into their languages). FWIW I haven’t checked the quality of the articles either Kowal2701 (talk, contribs) 15:58, 11 June 2026 (UTC)[]

As such, we should have latitude for experiments from users with a deep track record who steer clear of poor-quality copy/pasted LLM-generated hallucinations. No, we shouldn't. Wikipedia is an encyclopedia, not a platform for text generation experiments. A "deep track record" (thousands of edits, multiple GA's, whatever) isn't a get-out-of-jail-free card for writing undisclosed advertisements for pay or violating copyright; it shouldn't be a factor in applying NEWLLM either. Stepwise Continuous Dysfunction (talk) 20:40, 12 June 2026 (UTC)[]

As such, we should have latitude for experiments from users with a deep track record who steer clear of poor-quality copy/pasted LLM-generated hallucinations. Wikipedia is not a laboratory for LLM experiments. SuperPianoMan9167 (talk) 23:43, 12 June 2026 (UTC)[]

Woah woah woah, is that an AI-generated comment? Not only is it quite wrong (per everyone above), but also quite long. Please don't write an entire essay just for a single comment. | One Reaction was here. Got a complaint? 13:29, 15 June 2026 (UTC)[]

No, it's not AI generated, and it's 300 words, not a wall of text. ScottishFinnishRadish (talk) 13:33, 15 June 2026 (UTC)[]

Thanks for clarifyng not AI, but i'm not sure about not wall of text. | One Reaction was here. Got a complaint? 13:43, 15 June 2026 (UTC)[]

@OneReaction5890, this is very clearly not an AI-generated comment. Please do not make such accusations on a whim, more certainty is needed Kowal2701 (talk, contribs) 14:08, 15 June 2026 (UTC)[]

It is with great regret that I rescind I'm willing to assume on good faith that you know what you're doing, as I was convinced to check more deeply in this case, and the issues do NOT look pretty, aside from bucketloads of WP:AIATTR, WP:AILIST, and other more vague style issues:

• He Tingbo:
  • The citation [119] returns a 404. The actual url is at [120]. No archives found at the first url.
  • Sources 10 and 11 have the same title. Bloomberg is the correct one [121]; Fortune returns an error page.
  • Sources 1 and 3 are cited to the statement In 2003, Huawei founder Ren Zhengfei placed her in charge of the company's chip development, providing an annual budget of US$400 million and a mandate to reduce dependence on foreign semiconductor suppliers. The numbers are supported by Reuters, but "a mandate to reduce dependence on foreign semiconductor suppliers" is not explicitly found in either source. In fact my read of the Reuters source is that the mandate was not originally that, but eventually became that, with the associated consequences.
  • In the years following the US restrictions, He led HiSilicon's effort to redesign its chip portfolio using alternative approaches that did not depend on restricted manufacturing technologies is sourced to [122], which mentions neither HiSilicon nor US restrictions, and Nikkei Asia, which does mention both (checked through university subscription), but is mostly on He's role inside Huawei rather than any specific HiSilicon effort to redesign anything.
• Joe Ando:
  • [123] is cited for He later played the recurring role of Rodney in the ninth and tenth seasons of The Walking Dead. but does not mention the ninth and tenth seasons.
• Irati Gorostidi Agirretxe:
  • [124] is cited for Agirretxe being awarded the Fulbright scholarship, which is not in the source, and and I cannot find Agirretxe in the Fulbright scholar directory.
• Steve Sweeney (journalist):
  • Source 2 does not say Sweeney was "the international editor" of the Morning Star. Using the article list page to surmise what he reported on is a bit SYNTH, but is relatively minor.
  • The Israel Defense Forces (IDF) claimed the strike targeted military infrastructure and that explicit warnings had been issued to civilians in the area. is cited to [125]. No Israeli response is mentioned in the article.
• Six AI tigers:
  • [126] (source 2) is a bad domain; there is no "ftp" in front in the real article.
  • [127] cites author Rita Liao; it is not clear where this is found in the article itself.
  • [128] has a publish date of 13 March; 15 May is the update date. Also why is the ref name Reuters2025?
  • All six companies reached "unicorn" status (valuations exceeding US$1 billion) by early 2024 (source 3) only mentioned Zhipu raising a billion in 2025, and does not mention any other company.
• Shawn McCreesh:
  • [129]'s author is Ariana Ortiz, not Malak Kassem.
  • [130] does not say that McCreesh was raised in Hatboro, only that he wrote something about it.
  • [131] does not say that he was involved with The Torch.
  • [132] does not confirm that McCreesh is the WH correspondent of NYT. I cannot in fact find any evidence that McCreesh is a WH correspondent for the NYT.
• Fight Unlawful Conduct and Keep Individuals and Communities Empowered Act:
  • Source 1 has the wrong author field. Source 2 has the wrong title (took the abbreviated URL; real title is NJ panel advances bills to strengthen immigrant protections, target ICE).
  • The legislation was introduced as part of a broader package of bills aimed at increasing state-level protections for immigrants following a series of high-profile enforcement actions by U.S. Immigration and Customs Enforcement (ICE) in Hudson County. for Source 2 - the source does not mention Hudson County.

I have more issues saved from my various automated tools, but that's a long enough post that demonstrates a regular pattern of problematic edits. Any new editor faced with this edit history would receive a swift block followed by equally swift LLMPROD. @Fuzheado, it would be appreciated if you could explain yourself. Fermiboson (talk) 14:33, 17 June 2026 (UTC)[]

Fuzheado thanks for including clear edit summaries. I do think people working with AI tools for analysis, review, research, or translation sometimes have non-overlapping workflows with those focused on AI cleanup. And while some languages-projects (like ko) have an active community of use around a particular constructive toolkit that allows for the development of positive policies, we don't quite have that here.

For people using LLMs anywhere in their research / planning / review workflows, it is useful to capture the workflow in a public + collaborative space, including prompts and context and workflows. That allows for collaboration, improvements over time, and catching errors or bias as early in the workflow as possible, not just in the final outputs. (A noticeboard for "AI misuse" is probably not the place. But considering the range of LLM tools used in [re]search by on the internet, we do need a place to discuss such things.) Current guidelines may discourage some from doing this, which isn't an ideal equilibrium. – SJ + 03:46, 16 June 2026 (UTC)[]

ScandalousSkink (talk · contribs · deleted contribs · logs · filter log · block user · block log)

ScandalousSkink started editing about one hour prior to this report. In that time, they have 29 edits, many of which are multi-thousands of characters long, and have markdown. I have not found spurious links yet or other obvious signs of unreviewed AI usage, but the pattern of activity is very suspicious. They have responded to other notices regarding their editing, but not to the AI notices. — rsjaffe 🗣️ 18:48, 10 June 2026 (UTC)[]

I have replied to you despite your claim otherwise. I used a Python script and these content was written in advanced. Zero LLM use and every edit factual and neutral. ScandalousSkink (talk) 18:51, 10 June 2026 (UTC)[]

Could you provide the code of the Python script? Gnomingstuff (talk) 20:01, 10 June 2026 (UTC)[]

Changing my mind on not participating on this. I personally don't think AI is being used here, but SS's unreverted edits do need cleanup. The big problem is that they're editing on mobile, which means that if they click the normal edit button, they only edit the lead, you need to open the options and choose "Edit full page" to get the rest of the article. Since they're pasting full article rewrites into the lead, the rest of the old article is still present.

Those pages need some cleanup, but it seems like there's quite a bit of useful material being added. At least one rewrite seems to be a marked improvement from the previous version, but I feel out of my depth out of my depth with checking the rest of their edits. ᴸᵃᶠᶠʸTaffer💬(they/she) In solidarity 19:17, 10 June 2026 (UTC)[]

My apologies to ScandalousSkink regarding asserting that they were non-responsive, as they responded to my caution about AI usage while I was working on my posting here. — rsjaffe 🗣️ 20:07, 10 June 2026 (UTC)[]

Some AI signs at Controversies surrounding Mahatma Gandhi, noted on my talk page: User talk:Asukite/Archives/2026/June#Controversies surrounding Mahatma Gandhi. User has admitted to using Grammarly in the past. A second look would be appreciated. ASUKITE 18:55, 10 June 2026 (UTC)[]

I took a look at Omniking2100 (talk · contribs) edits to Xhosa language. Placeholder citations in [133]:

• AuthorLastName, AuthorFirstName (Year). Title of Linguistic Study or History. Publisher/Journal. URL or DOI if available.
• AuthorLastName, AuthorFirstName (Year). Study on Khoisan Influence in Bantu Languages. Journal Name. URL.
• AuthorLastName, AuthorFirstName (Year). Clarifying Misconceptions on Xhosa Linguistics. Research Publisher.
• AuthorLastName, AuthorFirstName (Year). Oral Traditions in Xhosa Culture. Publisher.
• Dugmore, Henry Hare (1859). The Xhosa Bible. Publisher.
• AuthorLastName, AuthorFirstName (Year). Multilingualism and Language Change in South African Urban Centers. Linguistics Today. URL.
• AuthorLastName, AuthorFirstName (Year). Contemporary Studies in IsiXhosa Linguistics. Academic Press.

This is already very telling. This edit, and also [134] and [135] show WP:AILEGACY, WP:SUPERFICIAL, WP:AICURLY, WP:AIWEASEL, WP:AIPARALLEL, WP:AITITLECASE and some other oddities.

Their edits are clearly LLM assisted. MeowsyCat99 (they/them) (meow) 20:39, 10 June 2026 (UTC)[]

Other than the stylistic flags. On the placeholder citations I'm not sure what this proves against, I understand it's Ai, but exactly what it means if I have a different placeholder then and now. Also last month I didn't even know what a SYNTH was or a citation template. Those edits were made by a different version of me, a less knowledgeable one. I was a casual, editor to fix and improve an article about a topic I was biased to. Did I overly rely on Google for those inputs, yes, literally when you don't know you go to Google. Did I write this stuff as how Google instructed me to do it, yes definitely. I'm just trying to give you an idea of what I was probably doing. I don't really remember that much ngl, cause it was random days, I was reverted many times even on the worldbox article . Also there was a point I didn't know you had to put actually sources or url, I thought it was general knowledge, so I would go to a source and see ft says this and be done. But I definitely never used Ai. Also I get you're trying to build on possible connections, but this ain't really helping, look at my misconceptions article or my Elon before the afd, and current Elon totally different things, more policy based. So something from 5 months is obviously gonna be problematic. That's why I like when it's about current topic this article and see where is the fault. Omniking2100 (talk) 21:38, 10 June 2026 (UTC)[]

What exactly do you mean by I understand it's Ai, but exactly what it means if I have a different placeholder then and now.?

You say: Did I overly rely on Google for those inputs, yes, literally when you don't know you go to Google.. Can you please describe what tools you use when editing? By "Google", do you mean Google search, or Gemini? Do you use any tools in addition to Grammarly?

You say: Did I write this stuff as how Google instructed me to do it, yes definitely. What do you mean by Google "instructing" you? MeowsyCat99 (they/them) (meow) 21:49, 10 June 2026 (UTC)[]

Oh no I'm just saying I understand the context is an AI accusation, that's what I meant by that. On the Google thing, when you search how to do citations, Google gives you an answer on its own, not just a list of pages. It shows you examples and templates like `{{cite web |url= |title= |website= |publisher= |access-date=}}` and you take that, save it, and use it as a template you fill in whenever you need a citation. That's what I mean by google instructing me. I'm not talking about gemini or any chat ai, I mean the search engine giving you structured answers. Back then I was only on phone, no laptop, those article fixes were on my phone made by phone, no draft or user sandbox, just write and shove in there. I'm hyper aware the Ai gemini is integrated into google , when you search there's that AI overview that pops up automatically. But I never went into it to chat or respond atleast not in the context of stuff for Wikipedia. You just take what's in front of you and draft yourself. It's policy based cause it's not writing anything for you, it's just showing you how things are. Grammarly is the interesting one I hope you read my earlier argument on its use it was that link I I gave the user who brought this debate. I think Wikipedia policy hasn't fully considered what it is yet, because it actually meets the criteria for acceptable use in certain ways. So even the policy is unclear on it specifically.On tools I genuinely don't know what tools you mean. The only tool I know is AI, which I've been saying I don't use. Ive seen Wikipedia mention automated tools but never really understood what those were. If you mean those, I've never used them either in fact the other user just told me he used a tool to auto clean the citations, I can barely put correct links in my articles to articles that talk about them cause it's hard manually. You can actually test this yourself search Google for how to make a Wikipedia citation template and see exactly what it gives you, it will be like how it gave me back then, it sometimes even gives you an easy to copy so you click that button and the example is sent to you clipboard. I do really wanna say I remember I contributed to those articles, but I'm just telling what I I think I remember I did, obviously I know I didn't use Ai, but I'm just trying to say I don't remember much of that ngl. Omniking2100 (talk) 22:32, 10 June 2026 (UTC)[]

A few snippets from your talk, which I think will speak for themselves:

• I have never used Ai. On the searching for a citation, ngl, on my drafts, before I even put a word, I must find a source that will talk exactly what I'm taking so I can get points there. So I do use claude Ai, chatgpt hallucinates and it's useless, but claude does go.
• other Ai I did use like claude or perplexity were for searching the right news citations and even that I was the one who went to the site and took the link and wrote the stuff.
• That's how it works no Ai, I mean yes Ai in that specific one. I do remember I was told this previously on my Elon I had used Ai and nervous it was my first page and stuff and tried to explain never really understood of it work or not never got a reply, cause that page has been there for months and I added the citations and checked and stuff. Idk what constitute as using Ai, do I use Ai to discuss the certain topic yes, do I use grammarly to auto correct my words yes, it doesn't improve them corrects

Those may have been taken out of context, but if anybody wants to verify that I will applaud them for finding the context. See also: WP:WALLOFTEXT - it would be helpful if you could focus your replies on a few points, and keep them concise. ASUKITE 22:55, 10 June 2026 (UTC)[]

On the policy WP:NOLLM it explicitly permits LLMs for basic copyedits like spelling and punctuation provided no new content is introduced. That's literally all I've used Grammarly for, individual word corrections. So that accusation is baseless, I've been clear on this repeatedly. I'm glad you acknowledge the context issue because that's exactly it. The previous user's questions were about articles from months ago — what you pulled is 30 days younger. First time I ever mentioned Grammarly was in April , not in those earlier edits. What I said in detailed defenses of separate articles is completely irrelevant to this one. Quoting isolated lines without context doesn't prove anything. Clarify what you mean it's "obvious". Goalposting is how I lost my previous articles in these discussions, I'd really appreciate it if we can be specific on this article I was called in for. Omniking2100 (talk) 00:13, 11 June 2026 (UTC)[]

I appreciate this reply. It is much more to the point than your previous ones, to the extent that it looks like it was written by somebody other than you, which may just be because you took the time to boil it down.

This is not just a discussion about one article, but about whether or not you are using LLMs to generate article text. Using them for sourcing can be problematic too, but I will admit I have checked every source link on the Ghandi article, and while I think some reuse the same URLs, they do all seem to link properly.

I posted the items that were the most obvious to me already on my talk page. I've spent enough of my time on this, and I don't want to seem like I'm badgering you, so if another member of the community finds any valid points, they are welcome to add them, otherwise this discussion can close, the tag can be removed and you are free to go on. ASUKITE 00:40, 11 June 2026 (UTC)[]

Thanks, appreciate it. Omniking2100 (talk) 00:56, 11 June 2026 (UTC)[]

Can you please explain how you mistakenly included references with placeholder information in one your edits? MeowsyCat99 (they/them) (meow) 21:07, 11 June 2026 (UTC)[]

This a really vague question. Even if I consciously remember I did that, it would actually mean nothing. As I said very long ago, irrelevant to the current accusation given the context of my evolution. Also was very new, remember months apart from that, on my worldbox attempt at contribution I tried to fill in with Fandom sources or even what Wikipedia says its self. You can't expect much months earlier than that. Omniking2100 (talk) 07:01, 12 June 2026 (UTC)[]

So you don't know? MeowsyCat99 (they/them) (meow) 10:47, 12 June 2026 (UTC)[]

In this context, yes.I've already noted that was an inexperienced version of me. Those edits, of most have been reverted were a genuine mistake on my part. I fixed some, like in the Xhosa people initiaton. Some I don't find meaning they've been appropriately reverted. Omniking2100 (talk) 14:59, 12 June 2026 (UTC)[]

i have noticed that and i could see most of my edits has been reverted and i am working on it and i think it is because i am inexperience and learning more on it Kwasi Rex (talk) 10:45, 13 June 2026 (UTC)[]

Rapid creation of blatantly AI-generated drafts (with one erroneously accepted into mainspace). Consensus requested for WP:LLMPROD. Ca ^{talk to me!} 07:44, 11 June 2026 (UTC)[]

As far as I know, WP:LLMPROD only applies to articles, not drafts. Even WP:NOLLM seems to only apply in articles, so the rules preventing AI in draftspace are quite limited. I would support extending both policies to draftspace, if anyone were to propose them, but the best option at this point is just to warn the user and wait 6 months. If they continue creating these AI pages, they could be taken to MfD as tendentious submissions. Somepinkdude (talk | contribs), in solidarity 16:13, 11 June 2026 (UTC)[]

I have seen other editors argue that draft articles are material being incubated for main articles, and therefore WP:NOLLM should also apply there. Whether it actually does is a good question. I spot-checked Draft:Diasporiana and agree that it is LLM due to awful source/text discrepancies. M kuhner (talk) 17:16, 11 June 2026 (UTC)[]

I CSD'd this editor's draft (Draft:Chaoxing) as a hoax due to the sources making no mention of the subject. The second iteration is wholly different in just about every way, a complete rewrite, which is unusual for recreations, which often bear some semblance to the original. The original seemed to have more of a POV slant as well, such as To protect its commercial monopoly referring to the PDG format, while the new draft retains mention of the format, it doesn't mention monopoly. I will have to take a closer look at that one as the new draft seems to go into much more detail. ASUKITE 20:23, 11 June 2026 (UTC)[]

I would prefer AI drafts not be deleted early, they are useful for data gathering especially since a lot of newer AI patterns are not fully known, but if they get deleted the data can't get gathered Gnomingstuff (talk) 03:13, 12 June 2026 (UTC)[]

I see your point. On the other hand, if someone makes, let's say, a thousand LLM articles in draftspace, it's like the time several hundred gallons of water accumulated in the tent roof over a State Fair concert. It's not doing anything right now, but if it all comes down at once....SPLAT. Either an immediate splat in mainspace, or a horrible oozing one in AfC. (In real life, having seen what was building up, I was safely far away when it finally happened.) M kuhner (talk) 03:23, 12 June 2026 (UTC)[]

There are several thousand in draftspace at any given time. Gnomingstuff (talk) 10:19, 12 June 2026 (UTC)[]

Multiple sentences give away some of this article was most likely generated by a LLM. I may be wrong since i can't find if there was the revision wherein some user added AI-generated content. Nonetheless certain things like subjective descriptions ("unscrupulous and devious") probably should be cleaned up. HyacinthF (talk) 17:04, 14 June 2026 (UTC)[]

I don't think there are any obvious LLM signs here. The description of "naturally skilled but unscrupulous and devious" was present in June 2022, pre-ChatGPT, and there's been no significant changes to the text since then. Andrew Gray (talk) 17:47, 14 June 2026 (UTC)[]

However, @HyacinthF:, while I agree with Andrew Gray that it's not LLM, you're right to notice that it violates WP:NPOV and you should feel free to improve it. Things like Wriothesley happily lied to the council and betrayed his master in Wikivoice are not okay. M kuhner (talk) 21:16, 14 June 2026 (UTC)[]

Struck through that message because it's a bad idea to wordsmith this article. I just confirmed that it is copyright violation from a blog post and likely will have to be deleted entirely, so I don't want you to waste your time. M kuhner (talk) 21:27, 14 June 2026 (UTC)[]

I'd check the blog post against the article at the time the blog was written if I were you, given the article's long editing history. A history which would appear to make significant LLM use unlikely. And note that 'unscrupulous and devious', or very similar wording, has been in the article for over a decade. As for whether it should be, WP:NPOV doesn't outright rule out such phrasing if that is the consensus amongst sources.AndyTheGrump (talk) 21:34, 14 June 2026 (UTC)[]

You're right. Questionable material in article: 2015. Blog post: 2025. Surprising, though, because the blog post is much longer, covers a bunch of different people, yet has a reasonably consistent tone throughout. I suppose there could be a third source which both are copying, but it's also possible that the blog post is a mashup of Wikipedia articles.

I do not like the lack of neutrality throughout this material, but it's outside my area and I don't feel qualified to fix it.

Thanks for catching my error. M kuhner (talk) 21:50, 14 June 2026 (UTC)[]

As the article history clearly shows, it has been developed in small stages over 23 years or so. Do you think it plausible that the 125 contributors responsible have all been copying the same source? AndyTheGrump (talk) 22:06, 14 June 2026 (UTC)[]

The mashup theory does seem a lot more plausible. However the questionable material was all added by a single, now-blocked user in fairly large chunks in 2015, so I was concerned. Thanks for the better analysis, and I'll be more careful in future. M kuhner (talk) 22:10, 14 June 2026 (UTC)[]

There are a lot of cases on this page, some of them ongoing, others completed. Would an uninvolved editor (or several) please look through them and close the ones that are clearly finished? (I can't do most of them as I'm involved.) That will make it easier for us to see the ones that still need attention. You do not need to be an admin to make non-controversial closures: you can include {{subst:Nac}} in your closing statement to indicate it was a non-admin closure. Thanks! M kuhner (talk) 21:06, 14 June 2026 (UTC)[]

Archived a few. I wouldn't worry too much about involvement, very few of the reports here are consensus-seeking in a way where closure or archival requires evaluating consensus. If there's a report where cleanup is completed and no further action is needed, then feel free to close/archive it. However, I'd suggest waiting a week (give or take a couple days) in most cases unless it's very clear that there will be no further activity. fifteen thousand two hundred twenty four (talk) 10:21, 16 June 2026 (UTC)[]

Thank you! M kuhner (talk) 14:52, 16 June 2026 (UTC)[]

I was editing the Department of Defense Whistleblower Program article and using the visual editor, I saw a "potential AI-generated content" flag at the paragraph starting at "The DoD IG commits itself to ensuring..." and I agree with the flag, though I'm far too new to know how to rephrase it to make it sound less like an LLM, so I would like a more experienced editor to work on that. Ott883 (talk) 04:53, 15 June 2026 (UTC)[]

I can't seem to reproduce the flag in the visual editor, but the main problem with that paragraph is that it doesn't cite any sources and sounds a bit promotional. ("The DoD IG commits itself to ensuring..." according to whom, exactly?). I suspect the tonal problems will resolve if it is rewritten based on reliable sources that examine the DoD's whistleblower activities. Einsof (talk) 12:17, 15 June 2026 (UTC)[]

No citations? Promotional? Clear signs of AI. I'd say, tag the section (or the whole article) with {{AI generated}} and check through the whole article. | One Reaction was here. Got a complaint? 13:10, 15 June 2026 (UTC)[]

Deleting my previous response, because to my great surprise, we are collectively wrong about LLM in this article: there have been no large additions of text (except by bots fixing references) since 2013. So it really is a problem of writing, not LLM use. The section remains problematic on sourcing and tone, but apparently humanly so. M kuhner (talk) 16:03, 15 June 2026 (UTC)[]

how does one manage to type in such a way that they sound like an AI without using AI? | One Reaction was here. Got a complaint? 16:06, 15 June 2026 (UTC)[]

Syntactically, it doesn't really. A lot of awkward syntax, the only real "flags" are the use of "ensuring" and "fostering" multiple time, and those are not really used in the most common AI way. Gnomingstuff (talk) 17:03, 15 June 2026 (UTC)[]

I feel like the very pro-DOD slant of a lot of the article contributes to the sense that it's LLM. But after all, the LLMs learned this kind of nonsense from us. The bullet lists are displeasing, but that was the Powerpoint era and a lot of people learned how things should look from Powerpoint. Some bits are cribbed from government sources, which are often written rather badly and in a distinctive style that may set off peoples' LLM intuition (it certainly set off mine). M kuhner (talk) 18:58, 15 June 2026 (UTC)[]

• ChristinePgh (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Found this user while searching for placeholder URLs in references: Diff/1339147116. Lots of large rapid-fire edits and page creations. ‑‑gurkubondinn 13:13, 16 June 2026 (UTC)[]

[136] has markup Kowal2701 (talk, contribs) 14:55, 16 June 2026 (UTC)[]

And why does the edit summary say [e]xpanded stub? This is the first revision of a new page.. ‑‑gurkubondinn 15:00, 16 June 2026 (UTC)[]

not telling you anything you don't already know but the edit summaries are just pure chatbot responses. I like this one:

Important Strategic Insight This letter allows you to: Connect Fitzhugh to early frontier defense Show pre-war tension Strengthen your War of 1812 narrative Elevate Sodus beyond a minor settlement Gnomingstuff (talk) 22:54, 16 June 2026 (UTC)[]

I think that I've cleaned up everything that I could after this user, tagged the rest. Also obvious chatbot edit summary: Diff/1339354638 ‑‑gurkubondinn 09:57, 17 June 2026 (UTC)[]

• ReaderOfSciFiNovelsNPhilosophy (talk · contribs · deleted contribs · logs · filter log · block user · block log)

There is a discussion with the user at User talk:ReaderOfSciFiNovelsNPhilosophy § June 2026.

This is an OKA translator, has twice inserted OAICITE into articles. Has been making large edits that will be difficult to check. ‑‑gurkubondinn 13:33, 16 June 2026 (UTC)[]

Another OKA translator? Keep seeing them everywhere Fortek67 (talk) 16:41, 16 June 2026 (UTC)[]

Remind me again why we allowed them... did people actually think that anything good could come of the combination of "AI" and "editing for pay"? Stepwise Continuous Dysfunction (talk) 19:21, 16 June 2026 (UTC)[]

IMO it's not a bad idea but users should be experienced Wikipedia editors. I've previously rejected AfC submissions from these editors, because after heavily reviewing them, these articles are poorly reviewed by the editors. I think OKA editors should be fluent in both English and the translating language, be familiar with the topic, and properly review these articles. Fortek67 (talk) 19:29, 16 June 2026 (UTC)[]

Looking at Special:Contributions/7804j does not exactly inspire confidence either. Created 19 new articles two days ago, sometimes as little as two minutes apart. ‑‑gurkubondinn 09:07, 17 June 2026 (UTC)[]

the perverse incentive strikes again Gnomingstuff (talk) 22:53, 16 June 2026 (UTC)[]

Tried to point that out in User talk:ReaderOfSciFiNovelsNPhilosophy § Disambiguation link notification for June 16, but no response yet. I think it's a bit much to pay someone to add 162 sources that they (presumably, by looking at the timestamps) haven't read and then expecting other people to read and verify them for free. ‑‑gurkubondinn 09:04, 17 June 2026 (UTC)[]

Don't ask me, I agreed with the suggestion that OKA editors would be "prohibited from directly editing articles entirely" (in the capacity of inserting paid machine translations on behalf of OKA). ‑‑gurkubondinn 16:48, 17 June 2026 (UTC)[]

Continually removing AI tags in rather hostile fashion, claiming they are not using AI. Edits like Special:Diff/1223722511 and Special:Diff/1222630998, I feel, speak for themselves. (Also look at timestamps; for instance, the latter was added 3 minutes after the previous edit.)

I do not think that all of the text is AI-generated; for instance, the last paragraph of Special:Diff/1235942581 not only doesn't sound like AI but has some URL parameters consistent with various Firefox search actions (it looks like this bug) that I'm not familiar with chatbots spoofing. It seems like the AI edits, for some reason, are interspersed in. Gnomingstuff (talk) 22:08, 17 June 2026 (UTC)[]