dictionary

Chain-of-Thought Prompting

Published Updated

Chain-of-thought (CoT) prompting mimics human problem-solving by asking a model to show its intermediate reasoning before stating a conclusion. Instead of jumping directly to an answer, the model is guided—either through examples or explicit instructions like "think step by step"—to decompose the problem into logical steps. This substantially improves performance on arithmetic, logic, and multi-step reasoning tasks.

CategoryPractices
Reading time4 min read

Definition

A prompting technique that instructs an AI model to reason step-by-step before arriving at a final answer, significantly improving accuracy on complex tasks.

Need this applied?

We help teams go from definitions to deployed workflows—safely and fast.

Chain-of-thought can be triggered in two ways: few-shot CoT includes example problems with their step-by-step solutions, teaching the model to mimic that format. Zero-shot CoT uses a simple suffix like "Let's think step by step" to activate the same behavior without examples.

  • • Few-shot CoT: provide worked examples showing reasoning steps in the prompt.
  • • Zero-shot CoT: append "Let's think step by step" to elicit reasoning without examples.
  • • Self-consistency: sample multiple reasoning chains and take the majority answer for higher accuracy.

When to apply it

Wei et al., 2022

Chain-of-thought is most valuable for tasks requiring multi-step logic: math problems, strategic planning, code debugging, and legal or financial analysis. For simple retrieval or classification tasks, CoT adds latency and cost with minimal benefit.

FAQ

Does CoT work on all models?

Chain-of-thought prompting is most effective on larger models (generally 7B+ parameters). Smaller models may produce incoherent or circular reasoning chains rather than genuinely helpful steps.

What is the difference between CoT and ReAct?

Chain-of-thought is purely about reasoning in text. ReAct (Reason + Act) extends CoT by interleaving reasoning steps with real actions like tool calls or web searches, enabling agents to gather new information during their reasoning process.

Email this summary + checklist

Get a copy of “Chain-of-Thought Prompting” and an AI readiness checklist in your inbox.