AI glossary
Chain of thought
Having a model work through a problem step by step before giving the final answer, which improves accuracy on reasoning tasks.
Early research showed that asking a model to “think step by step” made it better at maths and logic. Modern reasoning models do this automatically: they spend extra tokens thinking before they answer, often hidden from the user or shown as a summary.
You can still help by asking for the steps that matter — for example, to list the assumptions or check the result against the original requirements.
Example: Asked “if a train leaves at 10:40 and the journey takes 2 h 35 min, when does it arrive?”, a model that answers straight away can get it wrong. With a chain of thought it adds the hours first, then the minutes, and reaches 13:15.
In practice
- With reasoning models you do not need to ask them to “think step by step”: good context is enough.
- If you want to see the reasoning, ask it to explain the key steps in the answer.
- The reasoning a model shows does not always reflect exactly how it reached the answer, so check the result anyway.
More ideas in our prompting techniques.
