{
  "dataset": "glossary",
  "record": {
    "id": "chain-of-thought",
    "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": {
      "value": null,
      "verify_status": "verify-against-primary-at-build",
      "source_hint": "https://en.wikipedia.org/wiki/Prompt_engineering — 'chain-of-thought prompting' named by Wei et al., Google, 2022"
    },
    "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"
  }
}