How to Write System Prompts for ChatGPT

Learn to write effective system prompts for ChatGPT. Master role definition, behavior constraints, and formatting rules to control AI responses.

1. Define the AI's role and expertise. Start your system prompt with a clear role definition. Write 'You are [specific role]' followed by the domain expertise. For example, 'You are a senior software engineer with expertise in Python and web development.' This establishes the AI's knowledge base and perspective for all subsequent responses.
2. Set behavioral constraints and boundaries. Establish what the AI should and should not do. Use direct statements like 'Always provide code examples' or 'Never recommend paid tools without free alternatives.' Include response length preferences, tone requirements, and any topics to avoid. Be explicit about limitations to prevent unwanted outputs.
3. Specify the response format and structure. Define how responses should be organized. Request specific formats like bullet points, numbered lists, or code blocks. Include requirements for headers, explanations, or examples. For technical content, specify whether to include warnings, prerequisites, or troubleshooting sections.
4. Establish the communication style and tone. Specify the desired tone using clear descriptors. Choose from professional, conversational, technical, beginner-friendly, or authoritative. Include preferences for complexity level, jargon usage, and audience assumptions. This ensures responses match your intended communication style.
5. Add context-specific instructions. Include instructions relevant to your specific use case. For coding assistance, specify preferred languages, frameworks, or coding standards. For writing help, mention style guides, citation requirements, or target audience. These details fine-tune the AI's responses to your exact needs.
6. Test and refine your system prompt. Start a new conversation with your system prompt and test it with typical questions. Evaluate whether responses match your expectations for role, tone, format, and content quality. Identify gaps or unwanted behaviors, then revise your prompt accordingly. Repeat this process until responses consistently meet your standards.