Why AI Gets Worse Mid-Conversation

If you've used ChatGPT for a while, you've probably noticed this. You start a conversation and the answers are great. Sharp, specific, remembers what you told it. Then twenty messages in it starts going generic. Forgets the tone you asked for. Contradicts itself. Starts sounding like it's answering a different question than the one you asked.

You didn't do anything wrong. The AI didn't break. It ran out of room.

What a context window is

Every AI conversation happens inside something called a context window. Think of it as a whiteboard. The model can only see what's on the whiteboard right now. Everything you type and everything it responds with takes up space on that whiteboard.

When it fills up, older stuff gets erased. Your original instructions, the background you gave at the start, the tone you asked for — gone. The AI at message 30 is working from a much narrower picture than the AI at message 5.

Why it fills up faster than you think

Pasting a long document burns through context fast. One wall of text can eat up as much space as a dozen back and forth exchanges. If you paste a contract, a report, a transcript — you're burning through your available window before the real work even starts.

Signs you've hit the limit

You'll know when it happens:

• It contradicts something it said earlier in the same conversation
• Answers get generic when they were specific before
• It forgets an instruction you gave at the start
• It starts repeating itself
• Responses start sounding like a generic FAQ instead of a reply to your actual question

What to do about it

Start a new conversation for each task. A new tab is free. You don't get extra credit for keeping everything in one marathon thread.

Paste less. Instead of dropping in an entire document, share only the section that's actually relevant to what you're asking right now.

Summarize before you restart. If you're partway through something and the quality starts slipping, ask the AI to compile everything important — decisions made, context, constraints — then paste that summary into a new conversation and keep going from there.

Once you understand what's happening, you stop blaming the tool and start managing the window. That's really all there is to it.

I teach this and a lot of other things people don't realize about how AI actually works over at Clearly, AI — a school built for everyday people, not developers. If you've ever felt like you're not getting as much out of these tools as you should be, that's probably where to start.