System Prompt

What a System Prompt Is

When you use a product built on an LLM — a customer service chatbot, a writing assistant, a coding tool — there are usually instructions that come before your first message. These are the system prompt: text sent to the model by the developer, separate from anything you type.

The system prompt shapes how the model behaves throughout the entire conversation. It's like a briefing you give someone before they start a job.

How It Differs From User Messages

In the API, there are typically three types of messages:

• System — set by the developer, invisible to the end user, sent before the conversation
• User — what the person using the product types
• Assistant — the model's responses

The system prompt persists across the whole conversation. User messages are part of the ongoing back-and-forth.

What System Prompts Are Used For

• Role assignment — "You are a helpful customer support agent for Acme Corp."
• Tone and style — "Always respond in a friendly, concise tone. Avoid jargon."
• Constraints — "Only answer questions about our product. Politely decline unrelated requests."
• Format requirements — "Always respond in JSON with these fields: answer, confidence."
• Safety rules — "Never give medical or legal advice."

An Example

A coding assistant might have a system prompt like: "You are an expert JavaScript developer helping a beginner learn to code. Explain your reasoning step by step, use simple language, and always include working code examples."

Every AI product you've interacted with almost certainly has a system prompt — you just don't see it.