Chain-of-thought (CoT)

Prompting a model to reason step by step before answering, which measurably improves accuracy on math and logic. Reasoning models now do this internally.