AI models have a fixed limit on how much text they can actively consider—the context window—so earlier points in a long conversation get crowded out, forcing you to either restart conversations or strategically reintroduce key details. Knowing this constraint helps you structure collaboration with AI more realistically, repeating context when needed rather than assuming it's retained.
Think of a context window like a whiteboard that an AI uses during a conversation. Everything you and the AI talk about gets written on this whiteboard. But whiteboards have edges. Once you've filled it up, the AI forgets what you wrote at the beginning to make room for new information.
Here's what's actually happening: every AI model can only "remember" a certain amount of text at one time. ChatGPT can hold roughly 8,000 words in its working memory. Claude can hold about 100,000 words. Gemini is somewhere in between. That sounds like a lot, but it includes everything—your questions, the AI's answers, your follow-ups, all of it.
Why you'd notice this: You're working on a long assignment. You paste in your assignment rubric, ask the AI some questions, have a conversation about your draft, ask for feedback—and then you ask a new question and the AI gives an answer that doesn't match what it said earlier. That's because the context window filled up and the earlier conversation got pushed out of the AI's active memory.
This especially matters for college work because assignments are inherently long. You need the rubric, the background material, your draft, and multiple rounds of feedback to matter. Losing early context means losing important constraints and details.
How to work around it: break big projects into smaller conversations. If your assignment has multiple parts, dedicate one chat to each part. Or, before the context window fills up, ask the AI to remind you of the key constraints: "Based on everything we've discussed, what are the three most important things I need to remember for the next draft?" It'll give you a summary you can paste into a fresh conversation later.
Different tools have different sizes, so Claude is generally better for long research projects, while ChatGPT is fine for shorter assignments or brainstorming.
Try this: If you're working on a paper longer than 5 pages, start a new chat after every major section. At the end of each chat, ask the AI to summarize what you decided about that section. Paste that summary into the next chat. You'll keep everything in alignment without confusing the AI.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.