The standing instruction at the top of a model's context that sets its role, rules, and behaviour for a whole conversation, given priority over user messages and usually hidden.
System Prompt is the standing instruction placed at the top of a model's context that sets its role, rules, and behaviour for a whole conversation, separate from the user's messages. It is where an assistant is told who it is, what it may and may not do, its tone, and any tools or formatting it should use.
The model reads it as part of the prompt, but it is given priority over ordinary user turns and usually stays hidden from the user. It persists across the exchange while user messages come and go, which is why it carries the durable rules rather than a single request. Attempts to override it from user or retrieved content are a form of prompt injection.
System prompts shape how a model answers questions about brands and products: the persona, the caution, and the instructions on citing sources are set here. Deliberately designing the surrounding instructions and context is context engineering; the system prompt is the most privileged part of that context.
