System Prompt

A system prompt is a set of instructions that you provide to an AI model before the conversation begins. It acts as a persistent directive that shapes how the AI behaves, responds, and formats its output throughout the entire interaction. Think of it as the AI's job description — it defines the role, the rules, and the expectations.

How system prompts work

Most AI platforms have two types of input:

• System prompt (or system message): Instructions that define the AI's behaviour. These persist throughout the conversation and are not typically visible to end users.
• User prompt: The actual question or request from the user.

The system prompt runs "underneath" every response. If you set the system prompt to "You are a senior financial analyst. Always cite data sources. Use British English," every response in that conversation will reflect those instructions.

What to include in a system prompt

Effective system prompts typically cover:

• Role: "You are a senior marketing strategist with 15 years of experience in B2B SaaS."
• Tone and style: "Write in a professional but approachable tone. Use short sentences. Avoid jargon."
• Format rules: "Always respond with bullet points first, then a detailed explanation. Include a summary at the end."
• Constraints: "Never make up statistics. If you do not know something, say so. Do not discuss competitors by name."
• Context: "The user is a product manager at a mid-size fintech company. They are evaluating AI tools for their customer support team."
• Output expectations: "Keep responses under 300 words unless asked for more detail."

System prompts vs regular prompts

The difference is persistence and priority:

• A regular prompt is a one-time instruction. "Write me a marketing email."
• A system prompt persists across all messages. "You are a marketing copywriter who writes in our brand voice. All emails should be under 150 words, use active voice, and end with a single call to action."

The system prompt sets the baseline behaviour. The user prompt provides the specific task.

Practical applications

System prompts are powerful for:

• Custom AI assistants: Build an AI that behaves like a specific expert for your team
• Brand voice consistency: Ensure all AI-generated content matches your brand guidelines
• Output standardisation: Get consistent formatting across different queries and users
• Safety and compliance: Set rules about what the AI should and should not discuss
• Domain specialisation: Focus the AI on your industry, products, or processes
• Workflow integration: Define the AI's role within a specific business process

System prompt best practices

• Be specific: Vague instructions produce vague behaviour. "Be helpful" is not useful. "Provide actionable recommendations with specific next steps" is.
• Set boundaries: Tell the AI what NOT to do as well as what to do.
• Include examples: Show the AI what good output looks like in your context.
• Test and iterate: Try your system prompt with various user inputs to ensure it handles edge cases.
• Keep it focused: A system prompt that tries to cover everything ends up covering nothing well.
• Update regularly: As your needs evolve, refine your system prompts accordingly.

System prompts in CLAUDE.md

For developers using Claude Code, the system prompt equivalent is the CLAUDE.md file — a persistent set of instructions that defines how Claude should work within your codebase. This is the same concept applied to development workflows.