Anthropic's Prompt Engineering Guide

Summary

Anthropic's own documentation on how to ask for what you actually want. It covers being explicit about the task, giving examples, assigning a role, and breaking long work into steps. The examples are readable without a coding background, and the patterns carry over to any assistant you use.

Key takeaways

• Specific instructions and examples outperform clever phrasing.
• Complex work gets better results when split into sequential steps.
• Asking the model to show its reasoning surfaces mistakes earlier.