Evidence-first authority page
What Is Chain-of-Thought Prompting?
A definition of chain-of-thought prompting, why it improves reasoning accuracy on certain tasks, and where it doesn't help.
Direct answer
Chain-of-thought prompting asks a model to show intermediate reasoning steps before giving a final answer, rather than jumping straight to a conclusion — this measurably improves accuracy on multi-step reasoning tasks (arithmetic, logic puzzles) but adds little benefit and extra cost on tasks that don't require sequential reasoning.
Last reviewed 2026-08-29
This page remains noindex while its page-level source receipts are completed and verified. It is available for review, not presented as publication-ready authority.
Why breaking down reasoning helps
A model generating a direct answer to a multi-step problem has to get every step right implicitly, in one pass, with no chance to catch an intermediate error. Asked to show its work step by step, the model effectively checks its own reasoning as it goes, and errors are more visible (to both the model and a reviewer) at the specific step where they occur rather than buried in a single wrong final answer.
Where it helps and where it doesn't
- Helps: arithmetic, multi-step logic, tasks with a clear sequential structure.
- Limited benefit: simple factual recall, classification, or tasks with no real intermediate reasoning to show — the extra tokens add cost and latency without improving accuracy.
- Some modern models perform this reasoning internally by default (sometimes marketed as 'reasoning models'), which changes when explicit chain-of-thought prompting is still worth adding versus redundant.
Practical implementation
The simplest version is adding a phrase like 'think step by step' or 'show your reasoning before answering' to the prompt. For agent pipelines, consider whether you actually need the reasoning trace in the final output or just the improved accuracy it produces — you can often ask for reasoning, then extract only the final structured answer for downstream use.