You would never hand a new hire a one-line instruction and expect strong work. Write something about our services. Summarize this. Make it better. You would tell them the goal, the audience, the constraints, and what done looks like. AI needs the same briefing. Most people skip it, then blame the tool for the weak result.

A good brief has five parts. You can hold all of them in your head once you have done it a few times.

  • The goal. What this output is for, and what decision or action it feeds.
  • The audience. Who reads it, and what they already know.
  • The constraints. Length, format, tone, and anything to avoid.
  • A model of good. One example of the kind of output you want.
  • The definition of done. What has to be true for this to be usable as-is.
Example

A small estate-planning law firm fed new client intake notes to AI and asked it to summarize the intake. It got back a tidy paragraph that buried the only three things the attorney needed.

So the attorney rewrote the request as a brief. This summary is read by the attorney in the five minutes before a first meeting. The goal is to decide whether the matter fits the firm. Flag asset complexity, any sign of a contested estate, and how urgent the timeline is. Five bullets, plain language, no legal restatement of what the client already said. The same notes and the same tool produced a summary the attorney could act on. The difference was the brief, not the model.

The make-it-better trap

Vague revision requests get vague revisions. Make it better tells the AI nothing about direction. Better how. Shorter. Warmer. More specific. Less hedged. Name the axis and the change lands. Cut this in half and lead with the cost is a brief. Make it pop is a guess, and the AI will guess back.

You already know how to do this

You brief people every week. You scope work for a paralegal, a junior associate, a new account manager. The skill transfers directly. The only new habit is doing it for the tool instead of assuming the tool will infer what a person would. It will not infer. It will fill the gaps with the most average answer, every time.

The takeaway

Before your next real request, write three lines. Who reads this. What it is for. What good looks like. Then prompt. Those three lines are the difference between a draft you rewrite and a draft you use.