Chain-of-Thought

In plain English

Chain-of-thought means getting the AI to work through a problem step by step before answering. Newer reasoning models do this on their own.

Analogy

Like asking someone to show their work on a math problem instead of blurting the answer.

Why it matters

Step-by-step reasoning improves multi-step logic, such as qualifying a lead against several conditions. On reasoning models with thinking turned on, the model already reasons internally, so you control depth with settings like effort. Anthropic describes manual chain-of-thought prompting as a fallback for when thinking is off.

In practice

"Think through each requirement, then state whether the lead qualifies."

Related terms

Sources

• Wei et al.: Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
• Anthropic prompting best practices: Leverage thinking