Chain-of-thought prompting is a technique where a model is asked to work through a problem step by step, writing out its intermediate reasoning before giving a final answer. For multi-step problems like math word problems or logic puzzles, this markedly improves accuracy compared with asking for the answer directly.
It was introduced by Wei et al. in the 2022 paper “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models.” The authors found that generating “a series of intermediate reasoning steps” significantly improves large models’ ability to perform complex reasoning. Importantly, the effect mainly appears in large models — smaller ones do not benefit much — making it an example of a capability that emerges with scale.
Chain-of-thought laid groundwork for the later “reasoning” models that are trained to think at length before answering.
Why business readers should care: Chain-of-thought is a simple, free way to make models more reliable on analytical tasks. It also underpins a whole class of premium “reasoning” models, so understanding it clarifies what you are paying extra for.