Explainer

Why is ChatGPT getting dumber? It's usually the conversation, not the model

ChatGPT starts sharp, then somewhere in a long chat it forgets things, repeats itself and drifts. That's rarely a secret downgrade. It's usually this one conversation wearing out. Here's how to tell, and how to fix it.

What's actually going on

When people say ChatGPT (or Claude, or any assistant) "got dumber", the cause is usually not a hidden model swap. It's that one long conversation has degraded. As a chat fills up, the model loses track of things you said early on, re-serves answers it already gave, and quietly drops rules you set. The common name for this is context rot: response quality tends to slip as the context window fills with more and more history.

The reason is mechanical. The model re-reads the entire thread on every reply. The bigger the thread, the more its attention is spread across a wall of old text — and the easier it is for the parts that matter (your constraints, your latest decision) to get buried under everything that came before.

The signs your conversation is cooked

You can usually feel it before you can name it. Watch for these:

It forgets rules you set. You told it something earlier in the thread and now it acts like you never did.
You keep repeating yourself. You've reminded it of the same constraint several times over.
It repeats itself. Answers come back nearly identical to ones you already got.
It contradicts settled decisions. Something you locked in earlier quietly comes undone.
It drifts off-topic. Replies get vaguer and less anchored to what you're actually doing.

How to tell the model apart from the thread

There's a simple test. Open a brand-new chat and ask it the same thing you're stuck on. If the fresh chat handles it well while this conversation keeps stumbling, the conversation is the problem — not the model. If the fresh chat struggles too, then the issue may genuinely be the task itself, and restarting won't help. This one check stops you from blaming the wrong thing.

Why "start a new chat" is the real fix

You can't un-rot a saturated conversation — there's no button that makes it forget its own bloat. The only reliable fix is to move your work into a fresh chat that starts clean and sharp. The trick is doing it without losing the decisions and context you built up. That means handing the new chat a tight briefing, not scrolling back through hundreds of messages.

How to restart without losing your work

Give the new chat a short handoff: the goal in one line, the decisions you already locked, the constraints the AI kept dropping (list these explicitly — they're the first to go), and the current state of the work. Paste it as the first message of a fresh chat, led by "pick up where we left off, don't re-ask anything below." A template:

Pick up exactly where we left off. Everything below is settled — don't re-ask it.

GOAL: [what we're producing, in one line]

LOCKED DECISIONS:
- [decision 1]
- [decision 2]

CONSTRAINTS (keep these every time):
- [constraint 1]
- [constraint 2]

WHERE WE LEFT OFF:
[paste the latest draft, or the exact question we were on]

Continue from here.

Two minutes of that, and you're working with a crisp assistant again instead of a tired one.

The catch with doing it by hand

Writing the handoff yourself works, but it's tedious — and people reliably forget the exact constraints they were most tired of repeating, which are the ones the new chat most needs. That's the job Uncook does for you.

The faster way

Move your work to a fresh chat with Uncook

Free & unlimited. No account.

Paste a share link to your ChatGPT or Claude conversation. Uncook reads the whole thing and writes a clean reboot prompt — your goal, locked decisions, and the constraints the model kept dropping, all carried across. Skim it, paste it into a fresh chat, and you're back to sharp.

Uncook my chat →

Honest about your data: a share link is fetched once through our server to read the conversation, and its content is sent once to our AI provider (Anthropic) to generate your handoff — then discarded, never stored, never used for training. A share link makes the chat viewable by anyone with the URL; un-share it once you're done.