There's a lot to love about NotebookLM, and I've seen it used for a variety of projects. I've even used it to suggest a workout routine for me and learn about topics like SBCs.

But my enthusiasm for the tool has decreased lately, as its newer features don't seem to achieve the same quality of its early success. This has made me concerned that it's going to lose the traits that made it special in the first place.

NotebookLM was the first generative AI tool that really impressed me

It broke through my skepticism

While I was initially excited by generative AI tools when the technology started growing, I often felt let down and underwhelmed by the real-life implementations. Chatbots frequently hallucinated; they were never as capable as the companies marketing them made them out to be, and they never actually made me more productive.

This turned me into a skeptic of generative AI. I would try out tools, but I never bought into the marketing hype. Even as colleagues raved about the usefulness of certain tools, generative AI never failed to elicit a raised eyebrow from me.

In many ways, this is still the case with plenty of tools. Gemini has gotten better, but it still hallucinates and gives incorrect information for simple queries like song suggestions. I ignore most of the AI features on my smartphone. While I have found uses for Copilot in Excel, the tool has left me generally underwhelmed.

NotebookLM bucked this trend. When I first tried the tool, I went in from the angle of someone very skeptical of its capabilities. While it wasn't perfect, it was miles ahead of anything else I had tried at the time.

Part of this was down to its accuracy. Because you set the sources for NotebookLM, it was a lot less likely to hallucinate. The information it provided was also of higher quality than that from generalist tools, if you used the right sources.

Meanwhile, its Audio Overviews left me impressed despite my initial doubt. Not only could it parse formal information and present it in an accessible way through its AI hosts, but the feature also allowed me to create audio summaries in other languages — including my second language, Afrikaans.

I started using it for a range of tasks, like summarizing information about health conditions to make them easier to understand for my family, and getting gaming tips for specific games I play. While NotebookLM is popular with students, you don't have to be one to appreciate its features. Anyone who wants to summarize information accurately from a range of sources can use the tool.

But newer features feel underpowered or bloated

The new research tools are particularly disappointing

Since my initial impression of NotebookLM was so positive, I couldn't wait to see what new features arrived. However, while there were some useful additions, there were also features that felt underbaked.

Featured notebooks, in particular, feel like bloat. I thought that these would be useful ways for people to access useful information, but they honestly feel more like advertising slots than anything else.

This feeling is increased by how few featured notebooks are available. I don't feel like I'm getting a broad overview of a topic from a variety of sources. I feel like I'm getting the viewpoint of a particular publication or author to promote their brand. Featured notebooks also can't be hidden, which is what really makes them feel like thinly veiled ads.

When the ability to search the web for sources and use the Deep Research tool came to NotebookLM, I was optimistic. But in reality, the features turned out to be disappointing. Firstly, the viewing tab for discovered sources doesn't make it easy to quickly vet the quality of the pages the AI has pulled for your notebook.

As a result, I would manually visit the links to see what pages the tool had suggested. The suggestions were disappointing at best, and troubling in the worst cases. Fast research tends to pull low-quality pages, while Deep Research pulls very academic pages. Deep Research also pulls a lot more sources, but it becomes difficult to check them due to the limited window available to scroll through and click on links.

This makes it difficult to find results that are both relevant and trustworthy. While researching wood veneers and refinishing furniture, the fast research tool added a safety data sheet for a specific brand of paint stripper. Safety information is important, but precautions often differ according to what product (and therefore ingredients) the stripper has. For me, this was a product I don't use (and that I hadn't mentioned in the prompt), so it was completely irrelevant as a source.

What was troubling, though, was the quality of sources when I searched for information about supplements for chronic migraines. Some of the sources were brochures without clear citations or dates, but the video that was pulled as one of the sources was from a YouTube influencer with no medical qualifications. This was also the only video used in the sources. However, manually performing the same search on YouTube ("supplements for chronic migraines") brings up videos from multiple doctors and medical foundations as the primary results.

There are other quirks I think NotebookLM should fix, like improving its video overview customization and the mobile app experience. But instead of polishing its existing features, the app is expanding with more underbaked elements that diminish its overall quality.

NotebookLM no longer feels like it's getting the focus it deserves

Google has other shiny toys to play with

I honestly wish that Google would refine existing NotebookLM features rather than trying to expand it with tools that aren't unique and don't work that well.

For example, NotebookLM's source limit affects what you can do with the tool. My colleague Mahnoor Faisal also points out that as the number of sources in a notebook increases, the chatbot struggles to produce reliable, high-quality output.

Even users on Reddit say that they've noticed a decline in the quality of the AI's responses. So, improvements are definitely needed.

Instead, many of NotebookLM's newer features overlap with other Google AI tools. Some feel like they're just avenues for getting more people to use Gemini features. Deep Research, for example, arrived on Gemini first.

While flashcards are currently limited to Gemini users who have a Workspace for Education account, you can generate quizzes in both NotebookLM and Gemini. Meanwhile, one of NotebookLM's flagship features, Audio Overviews, is essentially replicated in another Google Labs project — Illuminate. There are some differences between them, but it's getting more difficult to define what makes NotebookLM stand out.

I'm not giving up on NotebookLM yet

While I'm a bit disillusioned with NotebookLM's newer features, I still think it's a great tool. Currently, the big defining feature is the ability to set your own sources, which makes the tool extremely customizable.

However, I think it needs more quality-of-life features and refinements to remain relevant and competitive. And when it comes to new features, I hope Google ensures they enhance the NotebookLM experience and maintain the high quality that contributed to the tool's early success.