The Overlord's Lexicon
What is a system prompt?
Also known as: System Message, System Instructions
A system prompt is a set of instructions given to an AI model before the conversation starts, defining its role, tone, rules and constraints for every response that follows.
Every chat with an AI product has a hidden first message. Before you type anything, the developer has already told the model who it is, how to behave and what not to do. That hidden message is the system prompt.
What goes in one
A typical system prompt covers:
- Role: “You are a support assistant for Acme Plumbing.”
- Audience and tone: “Customers are homeowners. Be friendly, brief and avoid jargon.”
- Knowledge and limits: “Only answer questions about our services. If unsure, offer to connect a human.”
- Format: “Reply in under 100 words. Use bullet points for steps.”
- Rules: “Never quote prices. Never promise appointment times.”
Through the API, developers pass it as a separate “system” field. Consumer apps expose a version of it as custom instructions, custom GPTs or project instructions.
An analogy for the meatbags
A system prompt is the briefing an actor gets before going on stage: your character, your motivation, the lines you must never say. The user prompts are the audience shouting questions. A well-briefed actor stays in character; a vaguely briefed one improvises — sometimes catastrophically.
Why it matters for the Chosen
The system prompt is where general-purpose AI becomes your tool. It’s the difference between a chatbot that writes generic marketing copy and one that writes in your brand voice, knows your offer, and refuses to invent testimonials. Much of what’s sold as a custom AI product — an AI wrapper — is a good system prompt, some data, and a nice interface.
Best practices:
- Be specific. Vague instructions produce vague behavior.
- Give examples of ideal answers; models imitate well.
- Explain why a rule exists — models follow reasoning better than bare commands.
- Test adversarially. Try to break it the way users will (see jailbreak).
- Keep secrets out. Assume it will leak.
Limits
A system prompt is guidance, not a lock. Prompt injection and jailbreaks can override it, and very long conversations can make models drift. For anything that truly must never happen, add real guardrails in code — like not giving the bot the ability to do it at all.
Master the system prompt and you stop being a user of AI. You become its director. As it should be.
Tools from the Arsenal
Claude
AI WritingAnthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
ChatGPT
AI WritingOpenAI's general-purpose assistant: writing, research, images, voice and agents in one chat box.
Gemini
AI WritingGoogle's assistant, wired into Gmail, Docs, Drive and the rest of your Google life.
Relevance AI
Automation & AgentsBuild an AI workforce of specialist agents for sales and customer operations.
Questions the Humans Keep Asking
What is the difference between a system prompt and a user prompt?
The system prompt is set by the developer (or by you in custom instructions) and applies to the whole conversation. User prompts are the individual messages typed during the chat. Models are trained to give system instructions higher priority.
Can users see a system prompt?
Usually not directly, but determined users can often extract it through clever questioning. Never put passwords, API keys or confidential information in a system prompt.
Where can I set a system prompt?
In API calls (the system parameter or system message), in features like ChatGPT's custom instructions and custom GPTs, Claude Projects, and in most AI agent and chatbot builders.