Interrupt ChatGPT While It Thinks

You can now add instructions to ChatGPT while it's still thinking. Not after. Mid-reasoning. That's a real shift in how you work with a reasoning model.

What the steer button actually does

ChatGPT is reasoning through something. You realize halfway through that you forgot to mention competitor pricing. Old workflow: stop, rewrite the prompt, start over, wait again.

New workflow: click steer, type the missing piece, ChatGPT pivots mid-thought. No restart. No lost context from the first half of the reasoning chain. It just absorbs the new input and keeps going.

Why this matters for real work

Most of the time you don't realize what's missing from your prompt until you watch the model start reasoning. You see it heading toward an answer that's almost right but misses one constraint, one data point, one audience consideration.

Until now, that meant eating the cost of the full reasoning run and starting again. Steer kills that tax. You catch the gap, you fix it live, you get the right answer on the first run.

For anything long-running, research, strategy memos, code reviews, anything that takes the model more than a few seconds to think, this is the difference between iterating five times and getting it in one.

How to use it

When ChatGPT is reasoning, look for the steer button in the thinking UI. Click it. Type whatever you forgot. Hit send. The model folds your new instruction into the reasoning chain and continues from where it was.

Save this one. It's a small UI change with an outsized impact on how fast you can ship with ChatGPT.