Early AI chatbots were limited to text input only, with document upload becoming a feature after the initial wave of the chatbot boom. It’s usually still not the main selling point of any chatbot today, but it’s one of the most useful features for incorporating your existing work into an AI chat. Now, you can just drop in your documents and ask for whatever you need from it.
But not all AI models handle this equally well. Some break things down in a way that actually saves you time, while others skim the surface or miss what you were looking for. And it’s not just model-dependent either - it’s just as much about how you use the tool and what you use it for. ChatGPT, Gemini, and Claude aren’t created equal, they all excel at different tasks and have different weaknesses, which became evident after I used all three for a study session with the same documents...
Why ChatGPT, Gemini, and Claude?
And the documents I’m working with
The reason I often reach for conversational AI assistants over research assistants for some of my documents is simply because convenience beats specialization. Many of my study sessions happen in chatbots like Gemini, ChatGPT, and Claude because I can get information and explanations without having to provide much context or any sources. So using their upload and RAG abilities while I’m already in the interface, instead of hopping over to a tool like NotebookLM, is just more practical sometimes.
To test the best options for my documents, I standardized the setup as much as possible. I used the same files and prompts in each one - in my case, it’s a merged PDF file of some materials on the topic of human-centered design, and it’s around 30,000 words long but only 14MB in size. My prompts were designed to work across the entire document, not just to extract specific sections. I knew working with the free tiers of all three tools would mean running into limits, but that’s the whole point - I wanted to see which one could handle a large workload and help me understand the full content better.
My initial prompt is too long to paste here, but the instructions basically included:
- A summary
- Break the themes into sections
- A quiz plus feedback and learning guidance
- Formatting instructions
ChatGPT
Given its limits in the free tier, it wasn’t bad
Using ChatGPT’s free tier means running into the limit of only 3 documents per day, capped at 20MB each. This is definitely the biggest roadblock when it comes to using ChatGPT for studying with documents, along with the 8K-16K context window, which doesn’t expand with file uploads.
At first, ChatGPT pulled through with its responses. Out of all the bots, I probably like its conversational style the most as it keeps me engaged even with complex topics. It delivered on every part of my prompt, including a summary, sectioning, a quiz with immediate answers and explanations, and guidance on the next steps. It kept the key points easy to read, using a lot of structure with headers, bold letters, numbers, and bullets.
I ran into limits pretty quickly. After about 11 prompts, I noticed a drop in quality, shorter answers, and repetitive quiz questions that it had already asked me. Which means it probably switched me down to GPT-5.3 Mini or something else, and it definitely started losing context. So, while I love ChatGPT for its personal approach, it’s not the most reliable tool for long study sessions with my documents.
Gemini
It’s better suited for longer study sessions
I already knew Gemini offers more generous upload limits, with 10 files per chat, up to 100MB per file, and 32K tokens. But it also lacks ChatGPT’s creativity and conversational warmth, keeping the chat focused on the facts. So it’s a mixed bag. But I still got some great responses, especially when you enable Guided Learning, which is what I’ve done here.
It pulled more context from the file than ChatGPT did at once, but managed to present it to me in a shorter format, which made it quicker to get through. It also didn’t give me the answers to the quiz right away and wanted to gauge my knowledge first. This is a much better approach for actually learning and understanding a concept.
My session lasted about 45 minutes with around 15 prompts, and I never seemed to have actually hit the context ceiling. This makes Gemini much more suitable for real studying using the free tier - I just wish it had a more engaging tone because it felt a little impersonal.
Claude
A more balanced approach
Claude’s architecture is a little different from both ChatGPT and Gemini. The same 200K token limit applies to free and paid tiers, and instead, free users get a usage quota. Depending on your query lengths, you can get anywhere from 30 to 100 messages a day. Instead of getting them once-off, they’re spread across the day, depleting and replenishing every five hours. Keep in mind I’m using Claude for Desktop, not Claude Code.
I knew the quota was going to be my biggest roadblock since everything else about Claude’s free tier is generous - 20 files per chat up to 30MB each, and I’m using Sonnet 4.6. So I just had to be more intentional with my follow-up prompts so as not to waste them. I was instantly impressed with its response to my initial prompt. It hit a middle ground between ChatGPT and Gemini of being personable and engaging, but also factual and grounded.
I also much preferred the format it presented its responses in, which more closely matched what I described in my prompt - it even built little diagrams for me because it picked up on the fact that I’m a visual learner. More importantly, it maintained the entire context of the file at all times instead of handling it in chunks. It was also better at holding me accountable - whenever my follow-up queries diverged from the topic, it would gently guide me back to the learning module it had created for me.
The winner?
The clear winner for me was Claude. I’ve yet to encounter a bot that infers context as well as it does - it knows exactly what I'm looking for, presents all the information in easy-to-digest blocks, and gently guides me through every piece of the content. This is likely because it has the largest context window of them all.
ChatGPT is still a good option for sparring and brainstorming, it keeps things casual and engaging. If I had to pick an AI for making sense of information quickly, it would be ChatGPT. Gemini is great for longer and more focused study sessions, it’s a bit like working from a textbook. If I had to choose an AI to study for a big test from a curriculum, it would probably be Gemini.
Claude hits a nice middle ground; for self-guided learning with my own resources, Claude comes out on top. The only problem is its quota limit in the free tier, so you have to be very intentional with every prompt and can’t have many back-and-forths with it. It will be different for everyone depending on your use case.