System prompt vs user prompt
System prompt vs user prompt is the difference between a job description and a single request. A prompt is simply the text you give an AI. The system prompt is the standing instruction that tells the AI what its job, rules and tone are. The user prompt is what is asked this time.
Why the difference matters
Think of a new colleague. On the first day you explain the job once: what we do here, how we speak to customers, what is never allowed. After that you only hand over the task of the day. You do not repeat the house rules every morning.
An AI works the same way. If the rules live in the system prompt, nobody has to retype them, and nobody can forget one. The results stay the same from Monday to Friday and from one colleague to the next.
The two side by side
| System prompt | User prompt | |
|---|---|---|
| Written by | Whoever sets the AI up | Whoever is asking |
| Changes | Rarely | Every time |
| Contains | Job, rules, tone | Today's task and its details |
The AI receives both together. It treats the system prompt as the frame and the user prompt as the request inside that frame.
A worked example: one prompt split in two
A joinery firm wants help with follow-up emails on its quotes. At first, someone types the whole thing every time:
"You are a friendly sales assistant. Write a follow-up email for the quote to the kitchen studio, 12 oak doors, sent ten days ago, no reply. Keep it short, do not offer discounts, sign as the sales team."
It works. But the rules are retyped each time, and sooner or later one is left out. Split in two, it looks like this.
- System prompt. "You write follow-up emails for quotes from a joinery firm. Tone: friendly and direct. At most 90 words. Never offer a discount or promise a delivery date. Sign as the sales team."
- User prompt. "Customer: the kitchen studio. Quote: 12 oak doors. Sent: ten days ago. No reply so far."
Everything that is true for every quote moved to the system prompt. Only the facts of this one quote stayed in the user prompt.
Who writes the user prompt?
In a chat, a person types it. When an AI works on its own through your business records, the software can fill it in instead. It takes the customer, the quote and the date from the record and hands them to the AI. That is why the split is worth the effort: the hundredth email follows the same rules as the first.
What if the two disagree?
The AI gives the system prompt more weight. So "never offer a discount" normally holds, even if someone asks for one. But a system prompt guides the AI. It is not a lock. Rules that must hold in every case, such as who may see salaries, belong in the access rights of your software, not in a prompt.
Four checks for a good system prompt
- Does the first sentence name one job?
- Are the rules written as plain "always" and "never" lines?
- Does it say how long the answer should be and in what form?
- Could a new colleague read it and do the job?
Leave out facts that change, such as today's prices. They belong in the user prompt.
How Worknano handles this
In Worknano, an agent is an AI helper you set up for one job, and it keeps the two prompts in separate fields. The system prompt is the standing instruction that shapes the agent's role and tone. The user prompt adds the actual details each time the agent runs, such as a deal name or the text of a support ticket. You edit both in Config under Agents. The firm limits sit outside the prompts: an agent can only read what the signed-in user may read.
- Build your own agent, with a checklist for good instructions
- How to edit a page assistant and a reusable agent
- AI agents: prompts, inputs and outputs
Create a free workspace, open an agent and read its system prompt and user prompt.