← Back to Docs
Prompt Engineering

Chain-of-Thought Prompting

Force the model to reason step by step before committing to an answer.

The Recipe

Append a single phrase to your prompt that triggers explicit intermediate reasoning. The model will break complex problems into sub-steps, reducing errors on math, logic, and multi-hop tasks.

Let's think step by step.

Why It Works

  • Decomposes the problem into manageable sub-questions the model can solve individually.
  • Surfaces hidden assumptions and catches arithmetic mistakes before the final output.
  • Works across GPT-4, Claude, Gemini, and open-source models — no fine-tuning required.

Example

Prompt:

“A store sells apples for $2 each and oranges for $3 each. I buy 4 apples and 3 oranges with a $20 bill. How much change do I get? Let's think step by step.”

Model output:

4 apples × $2 = $8. 3 oranges × $3 = $9. Total = $17. Change = $20 − $17 = $3.

Variants

Zero-shot CoT

Just add the trigger phrase — no examples needed.

Few-shot CoT

Provide 2–3 worked examples with reasoning chains before the query.

Auto-CoT

Cluster diverse questions, generate chains automatically, prepend to prompt.