Here's a piece of advice that sounds wrong and isn't: your AI tutor gets worse the longer you talk to it. Not because the model degrades, but because of how these systems remember — and knowing when to walk away and start fresh is one of the highest-leverage habits you can build.
Why long chats go downhill
Every AI assistant has a context window — a limited working memory that holds the current conversation. Everything you've said, and everything it's said, has to fit inside it. As a chat gets very long, two things happen. Older detail starts to fall out of view, so the model effectively forgets things you told it near the start. And even within the window, a long, sprawling conversation gets harder for the model to navigate, so it starts to lose the thread — repeating points, contradicting things it said earlier, or drifting off the level you set.
People describe this as the model "getting dumber" over a session. It isn't getting dumber; it's running out of usable attention. The fix isn't a cleverer prompt. It's a fresh start.
The signs it's time
You'll learn to feel it, but the tells are consistent:
- It forgets something you clearly established earlier — your level, your goal, a constraint.
- Its answers get vaguer, or it starts repeating itself.
- It contradicts a point it made twenty messages ago.
- It stops pitching to the right level — suddenly too basic, or too advanced.
When two or three of those show up, the conversation is past its best. Don't fight it.
How to start over without losing your place
The obvious objection: "But we'd built up so much context — I don't want to lose where I'd got to." You don't have to. Before you leave a flagging chat, ask it for a handoff:
Summarise what we've covered and where I'd got to, so I can continue in a new chat.
You'll get a tidy paragraph or two. Open a fresh conversation, paste your original tutor prompt back in, then paste that summary underneath. You're back exactly where you were — same context, full working memory, sharp again. It takes fifteen seconds, and it's the single biggest quality upgrade most people are missing.
Two habits that make it painless
One chat per topic or session. Don't run a single endless thread for weeks. Start a new conversation when you switch subjects, or at the start of a new study session. It keeps every chat lean and focused.
Use a Project, if your assistant has one. ChatGPT and Claude both let you set up a Project with standing instructions. Put your tutor setup there and it persists across every chat automatically — so a fresh conversation is never a cold start.
This is exactly the kind of thing the Guide covers, because it isn't about any one subject — it's about getting more out of the tool itself. And every prompt in the library is designed to survive this: paste it into a fresh chat any time your tutor starts to fade.
FAQ
How long is "too long"? There's no fixed number — it depends on the model and how dense the conversation is. Watch for the signs above rather than counting messages.
Will I lose my progress if I start a new chat? Only if you don't grab a handoff summary first. Ask for one, paste it into the new chat, and you keep everything that mattered.
Isn't a bigger context window the answer? It helps, and windows keep growing — but even large ones degrade when stuffed full. A fresh, focused chat almost always beats a bloated one.