A concrete example
A system prompt might say: answer in English, cite your sources, ask for confirmation before any sensitive action.
Why it matters
It gives answers consistency and keeps certain risky behaviors in check.
You run into it in AI APIs, custom instructions, AGENTS.md, CLAUDE.md, or agent builders.
Don't mix it up with
Prompt engineering: Prompt engineering means phrasing a request with enough context, examples, and constraints to get a useful answer.
Guardrail: A guardrail is a safety rule that limits what an AI or an agent is allowed to do.
Common mistakes
- Filling it with contradictory rules.
- Believing it blocks every attack.
- Changing it with every message when it should stay stable.
Quick checklist
- First I check whether the word names a concept, a tool, a risk, or a metric.
- I tie it to a concrete case: A system prompt might say: answer in English, cite your sources, ask for confirmation before any sensitive action.
- I keep the main trap in mind: Filling it with contradictory rules.
Quick questions
What is System prompt in AI?
The system prompt is the top-level instruction that frames how the model behaves during the conversation.
Where will I run into System prompt?
You run into it in AI APIs, custom instructions, AGENTS.md, CLAUDE.md, or agent builders.
Which word should I read next?
Start with Prompt engineering, Guardrail, AGENTS.md.