System Prompt
The hidden instruction that shapes how an AI behaves before you say anything.
A system prompt is a set of instructions given to a model before the user conversation begins. It defines the model's persona, constraints, tone, output format, and scope of knowledge.
In the API, it appears as a message with role "system" (OpenAI) or the system parameter (Claude). In consumer products like Claude.ai or ChatGPT, it's invisible — already set by the product team.
What system prompts do:
- Set persona: "You are a concise technical assistant"
- Define constraints: "Only answer questions about cooking. Decline everything else."
- Specify format: "Always respond in JSON with fields: summary, steps, caveats"
- Inject context: "The user's account tier is Pro. Today's date is…"
Why it matters for practitioners: When building AI features, the system prompt is your primary control surface. A well-crafted system prompt reduces the need for per-request instructions, improves consistency, and can dramatically change output quality without changing the model.
System prompts are not foolproof — users can sometimes override them through prompt injection or jailbreaking techniques.
In plain terms
A job briefing before a contractor walks on site. The contractor (the model) hasn't met the client yet, but already knows: here's the project scope, here's how to communicate, here's what's off-limits.