---
title: System Prompts — Prompt Engineering Technique
description: A system prompt sets durable instructions — role, rules, format, and boundaries — that apply across the whole conversation. It governs behavior more reliably than instructions buried in a single user message.
canonical: https://prompeteer.ai/techniques/system-prompts
category: Output control & formatting
---

# System Prompts

A system prompt sets durable instructions — role, rules, format, and boundaries — that apply across the whole conversation. It governs behavior more reliably than instructions buried in a single user message.

## What it is

A system prompt is a special, high-priority instruction block that establishes the model's persistent behavior for a session: who it is, what rules it follows, what format it uses, and what it must not do. It sits above ordinary user turns in the message hierarchy, so its guidance persists as the conversation continues.

Providers treat system instructions as the right place for stable configuration — persona, safety and policy constraints, tone, and output conventions — while individual user messages carry the specific request. Using the system prompt keeps behavior consistent and harder for a single stray user turn to override.

## When to use

- You need behavior, tone, or rules that persist across every turn of a conversation.
- You are configuring an assistant, agent, or product with a stable persona and policies.
- Formatting or safety constraints must apply regardless of what the user asks.
- You want separation between fixed configuration and per-request input.

## How it works

1. Put durable configuration — role, rules, format, boundaries — in the system message.
2. Keep the specific request in user messages, not the system prompt.
3. Be explicit and prioritized; state the most important constraints clearly.
4. Test that user turns cannot easily override critical system rules.

## Illustrative structure

The structure separates configuration from request: a system message like "You are [role]. Always [rule]. Format answers as [format]. Never [boundary]." plus a user message carrying the actual task. The system block conditions every subsequent turn.

## Pitfalls

- Overloading the system prompt with rules can dilute the important ones.
- System instructions raise but do not guarantee immunity to prompt injection or jailbreaks.
- Contradictory system and user instructions produce unpredictable behavior.
- Very long system prompts consume context budget on every call.

## Sources

- [Anthropic — System prompts](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/system-prompts)
- [OpenAI — Prompt engineering guide](https://platform.openai.com/docs/guides/prompt-engineering)

## Related techniques

- [role-prompting](https://prompeteer.ai/techniques/role-prompting)
- [constraint-prompting](https://prompeteer.ai/techniques/constraint-prompting)
- [structured-output](https://prompeteer.ai/techniques/structured-output)