How to Write Better Prompts for Claude

Learn to craft precise prompts that get better responses from Claude. Master specificity, context, and structure for optimal AI results.

1. Start with a clear primary instruction. Begin your prompt with the core task. Write a single, direct sentence that states exactly what you want Claude to do. Example: "Write a product description for a mechanical keyboard" or "Explain how DNS resolution works." This becomes your anchor. Everything that follows builds on this foundation.
2. Provide relevant context and constraints. Add details that shape the response. Include the audience ("for a technical audience"), tone ("conversational, not academic"), length ("under 200 words"), and any domain-specific requirements. Context reduces ambiguity. If you're asking for code, specify the language and version. If you want an explanation, say whether you need beginner or advanced depth.
3. Request a specific output format. Tell Claude how to structure the response. Use phrases like "format as a numbered list," "use JSON," "provide a step-by-step guide," or "write in table format with three columns." Specify whether you want headers, bullet points, code blocks, or prose. Format requests eliminate guesswork and make responses immediately usable.
4. Include an example of desired output. Show Claude what good looks like. If you want a specific style of response, provide a sample. For writing tasks, share one completed example. For code, show the expected input and output. For analysis, provide a model breakdown. Examples are more powerful than descriptions—Claude learns from your format preference instantly.
5. Specify what to exclude or avoid. State what you don't want. Say "don't include disclaimers," "avoid marketing language," "don't use technical jargon," or "exclude external links." Negative constraints prevent Claude from defaulting to cautious or verbose patterns. This is especially useful when you want direct answers without hedging.
6. Evaluate the response and iterate. Send your prompt and review the output. If it's off target, identify what's missing. Did Claude misunderstand the scope? Was the tone wrong? Did it include unnecessary information? Revise your prompt with the new insight and resubmit. Each iteration teaches you what Claude needs to hear.
7. Use persistent system instructions for behavior across conversations. If you repeatedly need the same tone, format, or style, use Claude's Custom Instructions feature (available in claude.ai settings and via API system parameters). This ensures every conversation starts with your preferences already loaded. Define how Claude should behave by default without repeating constraints in every prompt.