Prompting· 2 min
Chain-of-thought prompting
Ask the model to 'think step by step' and it will often reason better, especially on math and logic.
Key idea
Explicit intermediate reasoning improves accuracy and makes errors easier to spot. Modern 'reasoning models' (o1, R1) bake this in.
Example
Instead of 'What is 17% of 243?', try 'Solve step by step: What is 17% of 243? Show your work.'
What to remember
- Works best on multi-step problems (math, code, planning)
- Reasoning models internalize CoT — you don't need to ask
- Self-consistency: sample multiple chains, vote on the answer