Chain of thought
Also called CoT
Letting the model reason out loud in steps before answering — it gets more right when it shows its work.
Think of it like
Doing long division on paper instead of trying to hold it all in your head.
Example
Nudging “let’s work through this step by step” measurably boosts hard math and logic problems.
How it actually works
Newer reasoning models do this internally, spending extra compute to think before replying. The visible scratchpad became a hidden one — same idea, just more of it, and often no longer shown to you.
For product teams
Letting the model reason step by step — better accuracy on hard tasks, now built into reasoning models.
For engineers
Intermediate reasoning tokens before the answer; boosts multi-step tasks, often internalized in reasoning models.
Related
- Prompt engineering — A staple of prompt engineering.
- Now baked into “reasoning” models by default.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome