Claude Prompting Techniques

Anthropic's free prompt engineering course covers a handful of techniques that consistently improve results with Claude. None are tricks, they are ways of being clearer about what you want.

Use structure to separate parts

When a prompt has distinct parts, like instructions, a document, and a question, marking them clearly helps the model keep them apart. Many people use simple XML-style tags for this with Claude.

<task>Summarize the changes for the changelog.</task>
<diff>
... the git diff ...
</diff>
<format>One bullet per user-facing change, present tense.</format>

Let it think before it answers

For harder problems, asking the model to reason step by step before giving the final answer improves accuracy. With reasoning-capable models you can also raise the thinking effort for tough tasks and lower it for simple ones.