A context window is the amount of text an AI can process in a single conversation, and it matters because hitting that limit mid-analysis means starting over or losing your train of thought. Understanding your model's window size upfront lets you plan whether you're working with single articles or full literature reviews in one session.
You're in a study session with ChatGPT. You ask 25 questions about Biology in one conversation. By question 24, the AI gives you an answer that contradicts what it said in question 3. This isn't stupidity—it's hitting a limit called the context window, which is how much information the AI can hold in mind at once.
Think of it like a tutor's working memory. A tutor can remember everything you discussed in one hour, but if you had a conversation with them, then stopped, then came back 10 hours later and had another conversation, they wouldn't remember the first one unless you reminded them. AI has a similar limit, except measured in tokens (roughly words) rather than time.
Every AI has a maximum context length—how many words it can consider at once. ChatGPT-3.5 has a context of about 4,000 words. ChatGPT-4 jumps to 8,000 or 128,000 depending on your version. Claude has up to 200,000 tokens. When you're in a conversation, the AI reads your entire conversation history plus your new message, then generates a response.
If your conversation history is very long, the AI is working with less "focus" on recent messages because it's processing lots of older information. Imagine reading a 20-page document before being asked a question about it—you'd probably forget details from page 3. AI has the same issue.
If you're in a 2-hour study session asking dozens of questions, the AI might lose context. Questions late in the session might not align with things you said early on. It's not that the AI forgot your statement—it's that considering 40 questions plus answers creates cognitive load.
The fix is simple: start a new conversation when a session gets long. If you've asked 20+ questions, start fresh. Paste your system prompt again if you use one. This resets the AI's focus.
Also, be explicit about context when needed. Instead of assuming the AI remembers that you're studying for a neuroscience exam, say it in your question: "Remember, I'm studying Neuroscience 101 for the midterm. Now, explain the blood-brain barrier..." This ensures the AI gives relevant-depth answers even in a long conversation.
Longer conversations don't just risk the AI forgetting context—they also slow down response times and sometimes cost more (some platforms charge per token). Starting fresh conversations isn't lazy; it's efficient.
Try this: Do a study session with one AI and track how many questions you ask before you start noticing inconsistencies or feel like responses are getting less tailored to your situation. Often it's 15-25 questions depending on the length of your questions and answers. Once you hit that number, open a new conversation. Notice how the first response in the new conversation feels more focused and relevant. That's context window optimization at work.
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.