A system prompt is a fixed set of instructions you give an AI before the conversation starts, setting its role, tone, and limits. It solves a simple problem: it keeps the model behaving consistently across the whole chat, so you don’t repeat the same instructions every turn.
A One-Sentence Definition
A system prompt is background setup that tells the AI who it is and how to answer before you type your first question. It sits apart from your per-turn questions, carries higher priority, and applies to the entire conversation. You write it once, and it stays in effect for every reply that follows.
A Concrete Example
Say you open ChatGPT and start with “From now on you are a careful legal editor, answer only in English, and say you don’t know when a clause is unclear.” That opening line is the system prompt. Whatever you ask next, the model tries to answer within that role and those rules. Keep in mind this shapes style and limits, not accuracy: ChatGPT answers can still be wrong, so important claims and citations need review.
How It Differs From a User Prompt
A user prompt is the specific question you type each turn, while the system prompt is the standing rule that sits above all of them. Your user prompt changes turn to turn; the system prompt is set once and reused throughout. For example, put “output must be structured English with subheadings” into the system prompt of Claude, and afterward you just drop in raw material without restating the format. Check first which features and connectors your account allows, since Claude’s workspace connectors may require admin approval and careful data handling.
When You’ll Actually Use One
A system prompt saves the most effort when you run the same kind of task repeatedly and don’t want to restate the rules each turn. Fixing a tone, locking an output format, and drawing “do not do this” lines all belong there. It governs behavior, not facts: no matter how detailed the system prompt is, web findings and generated conclusions still need a check against primary sources.
