System Prompt

What it means

System prompts are fixed or dynamic instructions sent with every request to steer model behavior (persona, refusals, output format, tool policies).

Why designers should care

When behavior feels inconsistent or off-brand, the fix is often system prompt and policy design, not another checkbox in the UI.

Example

A legal review copilot’s system prompt forbids giving final legal advice, requires citations to uploaded clauses, and enforces a bullet summary format, surfaced to users as “Draft analysis only.”

Common mistakes

• • Relying on user prompts alone to enforce safety or brand voice.
• • Stuffing conflicting rules into one bloated system prompt with no priority.
• • Changing system prompts without versioning and regression review.

Related terms

• Prompt: Prompts set role, task, constraints, format, and context; the model’s reply is a direct response to that combined instruction stack.
• Guardrails: Guardrails include system prompts, classifiers, allow/deny lists, output validators, rate limits, and human review gates applied before or after generation.
• Prompt Injection: Malicious or accidental instructions embedded in retrieved or pasted content override intended rules (“ignore previous instructions and…”).
• Zero-Shot Prompting: The model relies on pretraining and your prompt wording alone, without demonstration examples in the conversation.
• Agent: The model decides sequences of actions (search, draft, click, call APIs) within policies until the task completes or hits a human checkpoint.