Chain-of-thought prompting
Chain-of-thought is asking the model to think through the problem step by step before the final answer, and to put the answer in a separate tag so it is easy to extract later in the process. Anthropic presents this as a manual alternative when the model's built-in thinking capability is turned off. The reasoning makes the answer longer and increases the number of tokens.
On this page
Source content checked . English translation published 2026-10-27.
In business automation
In a problem that requires several steps of thinking, an immediate answer may skip steps. Chain-of-thought is asking the model to think through the problem step by step before the final answer, and to put the answer in a separate tag so it is easy to extract later in the process. Anthropic presents this as a manual alternative when the built-in thinking capability is turned off.
Examples
These examples illustrate a possible workflow. Adapt the connections and rules to your own systems.
Thinking before the final answer
In the prompt, the model is asked to lay out its reasoning step by step and to write the final answer inside a separate tag. The process that receives the output extracts only the content inside the tag and ignores the reasoning before it.
The reasoning makes the answer longer
A team notices that the answers are longer after it added a request for step-by-step thinking. The reasoning itself is counted in tokens, so in a process that runs many times it is worth deciding whether it is really needed.
Relevant tools
- Claude
- The definition comes from Anthropic's documentation, which presents it as a manual alternative when the built-in thinking is turned off.
What to check
- Forgetting that the reasoning makes the answer longer and increases the number of tokens.
- Not separating the final answer in a separate tag, which makes it hard to extract later in the process.