What it is

Asking the model to reason before answering.

Chain-of-thought (CoT) prompting tells the model to work through a problem step by step instead of jumping to an answer. The reasoning improves the result.