Chain-of-Thought (CoT)

Chain-of-thought prompting asks the model to work through its reasoning step by step before giving a final answer, rather than jumping straight to a conclusion. A simple 'think through this step by step' often dramatically improves accuracy on math, logic, and multi-step problems.

The reason it works is mechanical: as the model generates its reasoning, that text becomes part of its own context, and each subsequent token is predicted with that reasoning available. In effect, the model 'shows its work' and uses that work to reach a better answer.

Newer 'reasoning models' build this behavior in, doing extended internal deliberation before responding — so explicit chain-of-thought prompting matters less with them, but the principle is the same.