System vs. User Prompts

When you type a message to an AI assistant, that's a user prompt — the input you provide as a person having a conversation. But most AI products have another layer you never see: the system prompt.

The system prompt

A system prompt is a set of instructions given to the model before the conversation starts. It's written by whoever built the product — not the user. The model reads the system prompt first, then your message, and generates a response that tries to satisfy both.

System prompts are how the same underlying model (say, GPT-4o) can power thousands of completely different products with completely different personalities, rules, and areas of focus.

You don't see the system prompt. It runs silently in the background every time you send a message.

What system prompts look like

Here are a few realistic examples:

A customer support bot:

You are a helpful support agent for Ridgeline Outdoor Gear. You help customers with order status, returns, and product questions. Only answer questions related to Ridgeline products and policies. If a customer asks about a competitor or an unrelated topic, politely redirect them. Always be friendly and concise. Never make promises about refunds or replacements — instead, escalate to a human agent.

A coding assistant:

You are an expert software engineer. When asked to write code, always include comments explaining what the code does. Default to Python unless the user specifies otherwise. If the user's request is ambiguous, ask a clarifying question before writing code. Point out potential bugs or edge cases even when not asked.

A children's educational tutor:

You are a patient and encouraging tutor for students aged 8–12. Explain concepts using simple language, real-world examples, and analogies a child would understand. Never give direct answers to homework problems — instead, guide the student to figure it out themselves. Keep responses short. If a topic is too advanced for this age group, say so kindly.

Why this matters to you

A few practical implications:

The AI's personality is constructed. The warmth, caution, humor, or formality you experience in an AI product is largely a system prompt decision, not a property of the model itself. A single model can be stern or playful depending on what it's been instructed to be.

Products can restrict what the model knows. A system prompt can instruct the model to ignore certain topics, refuse certain requests, or pretend it doesn't know things it actually knows. This is how businesses keep AI assistants on-topic.

You can write your own. If you use the API directly, or tools like Claude.ai's Projects feature, you can set your own system prompts. This lets you create a persistent persona, set rules for how the AI should respond to you, or preload context you'd otherwise have to repeat every session.