AI & LLMs
What is a system prompt?
A system prompt is the standing instruction that sets an AI model's role, rules, and constraints for an entire conversation, separate from the individual requests a user sends.
New to Bolt? Sign up free.
Technical definition
A system prompt is a message supplied at the start of a model interaction that establishes persistent behavior: the assistant's role, tone, output format, scope, and prohibitions. It is passed in a distinct channel from user messages, and models are trained to weight it more heavily, though not absolutely.
A system prompt persists across turns without being repeated. A user prompt applies to one turn.
In plain terms
The system prompt is the job description. The user prompt is the individual task.
You write the job description once: who the AI is, what it should never do, how the answers should look. Every task after that inherits it.
Why it matters
Without a system prompt, an AI feature drifts. It answers off-topic questions, changes tone between replies, and gives away information it should not. Users notice.
If you are building a customer-facing assistant, the system prompt is where you set the boundary: this answers questions about our products, in this tone, and refers anything about refunds to a human.
One caveat worth knowing before you ship: a system prompt is a strong instruction, not a security control. It can be overridden through prompt injection. Anything that must not happen belongs in permissions and server-side checks, not in wording.
