When an AI draft comes back vague, the instinct is to rewrite the prompt. Usually the prompt was fine. What was missing was the brief — the small set of decisions a human writer would have made before opening the document.
Here is the one-page brief we ask for before generating anything. Six fields. None of them takes more than a sentence.
1. Reader
Not a persona document. One line naming who this is for and what they already know. "Marketing managers at 20–100 person software companies who already publish weekly" is useful. "B2B decision makers" is not, because it excludes nobody.
2. The one action
What should happen after someone reads it? Subscribe, book a call, change a process, share it internally. One action. Pieces with three actions reliably produce none.
3. The claim you are making
Every piece worth writing argues something a reasonable person could disagree with. Write that sentence down. If you cannot, you are writing a summary of a topic rather than a piece with a point, and no amount of prompting will fix that.
4. Facts that must appear
Numbers, product names, quotes, dates. List them explicitly with their source. This is the single highest-value field in the brief: it is what stops a draft from inventing a statistic that sounds plausible and is not.
A model will not tell you it is unsure. Give it the facts you want used, and check anything it produces that was not on your list.
5. Voice constraints
Three or four rules, negative ones work best. No exclamation marks. Never say "leverage". Product name always capitalised, never possessive. Short rules are followed; style guides are not.
6. What this is not
The most underrated field. "This is not a product announcement." "This is not a beginner's guide." Boundaries stop drift faster than instructions do, because they rule out whole directions instead of nudging.
Using the brief more than once
The point of writing this down is that it survives. The same six fields drive the article, the newsletter version, the social posts and the search snippet. That is why the derivatives sound like each other — they were all written from the same six decisions rather than from six separate memories of what the piece was about.
What still needs a human
The brief itself. Deciding who the reader is, what you are claiming and what you refuse to say is editorial judgement, and it is the part that determines whether the output is worth publishing. Everything downstream can be assisted. That part cannot.