Instead of returning an answer directly, chain-of-thought makes the model expose its reasoning process, for example by computing intermediate results. This helps especially with multi-step problems, logic puzzles, and arithmetic. The technique is often triggered by phrases like let's think step by step. Modern reasoning models use chain-of-thought internally and automatically, without needing to be asked.
Chain-of-Thought
Chain-of-thought is a prompting technique that asks the AI model to reason through its solution step by step. It noticeably improves accuracy on logical and mathematical tasks.