# Chain of Thought

> A prompting technique in which a model is elicited to produce intermediate reasoning steps before its final answer, improving accuracy on multi-step problems.

_The Agentic Web Lexicon · /glossary/chain-of-thought · [JSON](/api/glossary/chain-of-thought) · [all The Agentic Web Lexicon](/glossary)_

- **term:** Chain of Thought
- **category:** core
- **short_def:** A prompting technique in which a model is elicited to produce intermediate reasoning steps before its final answer, improving accuracy on multi-step problems.
- **long_def:** Chain-of-thought (CoT) makes the model 'show its work' — laying out the sub-steps of a calculation or deduction — which raises performance on arithmetic, logic and planning tasks. Modern reasoning models internalize this behavior, spending inference-time compute on a reasoning trace before answering.
- **see_also:** react-pattern, ai-agent, context-engineering
- **etymology_origin:** — (verify-against-primary-at-build)
- **related_to:** react-pattern, ai-agent, agentic-loop
- **contrast_with:** Unlike a direct answer, chain-of-thought first generates the reasoning steps, trading extra tokens for higher accuracy on problems that need more than one step.
- **example:** Asked a word problem, a model prompted for chain-of-thought writes out each step and only then states the total, catching errors a one-shot answer would miss.
- **source:** https://en.wikipedia.org/wiki/Prompt_engineering
- **status:** active
- **why_it_matters:** Reasoning traces are the substrate of agent planning; an agent that reasons step by step before acting makes more reliable tool calls on the agentic web.
- **sameAs:** https://en.wikipedia.org/wiki/Prompt_engineering
- **bridge_entity:** ai-agent
- **last_verified:** 2026-07-06
- **md_twin:** /glossary/chain-of-thought.md