On this page
Definition
Chain-of-thought (CoT) prompting is a technique where the model is asked to produce intermediate reasoning steps before its final answer, improving performance on multi-step problems.
Simple explanation
It is the difference between shouting an answer and showing your working. Asking the model to reason before concluding tends to reduce silly mistakes.
Why it matters
It unlocks better performance on maths, reasoning and planning without changing the model.
How it works
- 1Request reasoningAsk the model to 'think step by step'.
- 2Model works it outIt writes intermediate steps.
- 3Final answerIt commits to a conclusion based on the reasoning.
Real examples
Products named for illustration only. Inclusion is not an endorsement.
- Reasoning modesModern models expose 'reasoning' or 'thinking' modes that use CoT internally.
Advantages
- Improves accuracy on complex tasks.
- Makes model errors easier to spot.
Limitations
- Uses more tokens (higher cost, higher latency).
- Reasoning can itself be wrong.
Common misunderstandings
- ClaimThe model is genuinely reasoning.RealityIt is generating plausible reasoning-shaped text; sometimes that helps, sometimes it doesn't.
Frequently asked questions
Do I need CoT for every prompt?
No — for simple recall or transformation tasks it just wastes tokens.
The Tool Money Lab perspective
CoT was a genuine breakthrough. Modern 'reasoning' models formalise the idea, but the humble 'think step by step' still works surprisingly often.
Conclusion
Chain-of-thought turns quick guesses into worked-out answers. Use it when accuracy matters more than speed.