When should you use Chain-of-Thought (CoT) Prompting?

Multi-step arithmetic, logic puzzles, or word problems. Anywhere the model is making confident-but-wrong leaps. When the answer depends on intermediate calculations or sub-decisions. Reasoning about code: trace, simulate, debug.

When NOT to use Chain-of-Thought (CoT) Prompting?

Simple lookups or factual questions where reasoning steps add no value. Creative writing — explicit reasoning can flatten the prose. Token-sensitive contexts where cost matters and the task is easy.

How does Chain-of-Thought (CoT) Prompting work?

Models trained on instruction-following do better when forced to write intermediate steps before committing to a final answer. The act of generating reasoning tokens conditions the final answer on more deliberate context — like the model is talking itself through it. Variants: zero-shot CoT ("Let's think step by step."), few-shot CoT (give example reasoning chains in the prompt), or scaffolded CoT (require specific stages like restate → known → unknown → bridge → answer).

All techniques

Glossary · Technique

Chain-of-Thought (CoT) Prompting

Also known as: CoT, Let's think step by step

Force the model to think step-by-step before answering. Dramatically improves accuracy on multi-step problems.

Try the interactive template

When to use it

Multi-step arithmetic, logic puzzles, or word problems.
Anywhere the model is making confident-but-wrong leaps.
When the answer depends on intermediate calculations or sub-decisions.
Reasoning about code: trace, simulate, debug.

When not to use it

Simple lookups or factual questions where reasoning steps add no value.
Creative writing — explicit reasoning can flatten the prose.
Token-sensitive contexts where cost matters and the task is easy.

How it works

1Models trained on instruction-following do better when forced to write intermediate steps before committing to a final answer.
2The act of generating reasoning tokens conditions the final answer on more deliberate context — like the model is talking itself through it.
3Variants: zero-shot CoT ("Let's think step by step."), few-shot CoT (give example reasoning chains in the prompt), or scaffolded CoT (require specific stages like restate → known → unknown → bridge → answer).

Example

Lazy prompt

What's 17% of 250?

Using the technique

Solve step by step:
1. Convert 17% to a decimal.
2. Multiply by 250.
3. State the answer.
4. Verify by computing 10% + 7% separately and adding.

Common pitfalls

Reasoning can be plausible but wrong — verify with a different method when stakes are high.
Long chains burn tokens; if budget is tight, prefer scaffolded over open-ended chains.
On simple problems, CoT can introduce errors that wouldn't have happened with a direct answer.

Where this came from

Wei et al., 2022. Popularized by zero-shot CoT (Kojima et al., 2022).

Related techniques

Tree-of-Thoughts (ToT) Prompting

Generate multiple reasoning branches per step, evaluate each, and prune. Beats single-path Chain-of-Thought on hard decisions.

Self-Refine

Generate → critique own output → revise → repeat. Pushes a model's output much closer to its capability ceiling.

Try it interactively

The interactive template lets you fill in your scenario and generates a copy-ready prompt that uses this technique.

Open the template