Prompt basics for work: the five-line brief

Most "the AI got it wrong" stories are briefing problems. The model was asked for something vague and returned something vague. The fix is not a longer prompt. It is a prompt with the same five lines every time.

The five lines

1. Role. Who is doing the work. "You are an assistant to a regional sales manager."
2. Input. What it will receive, and in what shape. "You will receive the plain text of one email."
3. Output. Exactly what you want back, with the format spelled out. "Return at most three bullet points, each starting with a verb."
4. Rules. The things it must and must not do. "Do not invent deadlines. If no action is needed, return the single word NONE."
5. Example. One real input and the output you would have written yourself.

The example is the line people skip, and it is the one that moves quality the most.

A template to paste

Three habits

Ask for a fallback. Every brief should say what to return when the input does not fit ("NONE", "UNSURE", an empty list). Agents that always answer are the ones that make things up.

Keep the output machine-friendly. If a later step reads the result, ask for a fixed shape: one item per line, or a small JSON object. Prose is for people.

Version your brief. Copy it into a document with a date. When results drift, you will want to know what changed.

Try it now

Take one email from today. Write the five lines for "tell me if this needs a reply, and by when." Paste the email and read the answer. Then change one rule and run it again. That is the whole skill; the rest is practice.