The system prompts behind every major AI tool

What is a system prompt?

A system prompt is the natural-language instruction set an AI model receives before it ever sees a user's message. It is the model's briefing — its role, its rules, its personality, and the tools it is allowed to use. In modern large language models, the system prompt is a distinct message role (usually called system) that sits at the very top of the conversation and silently steers every response that follows.

System prompts are how a general-purpose model like Claude, GPT, or Gemini becomes a specific product. The same underlying model can be Cursor's coding agent, Perplexity's answer engine, or Notion's writing assistant — the difference is the system prompt. It tells the model who it is ("You are Cursor, an AI coding assistant..."), how to format output, when to ask clarifying questions, how to call tools, what topics to refuse, and what tone to strike.

A well-designed system prompt covers several concerns at once. It defines identity (name, role, voice), capabilities (which tools exist and how to call them), constraints (safety rules, refusal behavior, formatting requirements), workflow (how to plan, when to ask, how to verify), and context (environment info, current date, user preferences). Modern agent prompts can run thousands of tokens — Cursor's, Devin's, and v0's are each longer than most short stories.

Reading real system prompts is the fastest way to learn prompt engineering. Every prompt on this site is a production artifact from a team that has iterated hundreds of times on wording, ordering, and emphasis. You can see exactly how top AI companies handle edge cases, structure tool calls, prevent hallucinations, and shape agentic behavior — patterns you can adapt for your own AI applications.