How to Brief It

Most disappointing output is a briefing problem. That is good news, because briefing is a skill you already have — you use it with people — and it transfers across every AI tool you will ever touch.

The four things a good brief carries

Context. Who this is for, what they already believe, what has happened so far. Without it you get writing about the category of thing you asked for, which always reads as hollow.

A task. Specific and single. "Improve this" is not a task. "Cut this to 150 words without losing the guarantee" is.

Constraints. Length, tone, format, what to avoid. Constraints are not restrictions here; they are the most useful information you can give.

A definition of good. What would make you pleased with this? Say it.

A worked comparison

Write something about our bookkeeping service.

That produces filler, and it is not the model's fault — nobody could do better with that.

Write 150 words for a small business owner who has been doing their own books and is starting to fall behind. Plain English, no exclamation marks, no "in today's fast-paced world". Mention the free first month. End with one clear next step. It should sound like a competent friend, not a brochure.

Same model, same second. Entirely different output.

Show, do not only tell

When describing what you want is hard, show an example instead. Two or three samples of the kind of thing you are after carry information that no adjective can — rhythm, length, structure, the things you would struggle to articulate.

This is the fastest fix when a prompt is nearly working.

Say what to avoid

Models are much better at avoiding a named thing than at inferring taste. If output keeps doing something that grates, name it: no bullet points, no rhetorical questions, do not use the word "leverage".

Give it a role only if it means something

"You are an expert marketer" does very little. "You are a sceptical buyer who has been burned by an agency before" does a lot, because it constrains what the answer can be.

Iterate in one direction

When something is close but wrong, change one thing. Rewriting the whole prompt gets you something different-but-also-wrong and teaches you nothing.

Keep what works

The single highest-return habit: when a prompt produces something good, save it. Most people rewrite the same brief from scratch every time and never build anything.