Most content briefs are keyword briefs. A target phrase, a word count, three competitor URLs to imitate, a deadline. That process produces the fourth version of a page that already exists three times over, and answer engines have no reason to cite the fourth version of anything.
The brief is where a citation is won or lost. By the time a writer opens the document, the piece is already on its way to being quoted or already on its way to being ignored, and the difference is what the brief made the writer find and made the writer say. So we treat the brief as the deliverable that matters and the draft as its execution. Here is how we build one, in two halves: the research that finds the opening, and the one page that aims a writer at it.
Start from questions buyers actually ask
Skip the keyword tool for now. There are three better sources of real questions.
Sales calls first. The questions prospects ask in the first twenty minutes of discovery are the questions they asked an engine the night before, phrased almost identically. Pull the transcripts and write the questions down verbatim, wording intact.
Community threads second. Reddit, the industry Slack, the practitioner forums where your buyers talk when you are not in the room. The phrasing there is unguarded, which makes it exact.
The prompts you track third. If you run a set of buyer-shaped prompts against ChatGPT, Perplexity, Gemini, and Claude on a schedule, you already hold a list of the questions your market asks machines directly. If you do not track prompts yet, the sales calls will carry you for a quarter.
Notice what all three have in common: they surface questions with a decision behind them, not topics. "Does device fingerprinting still work after the iOS privacy changes" is a question. "Device fingerprinting trends" is a topic. Briefs built on topics produce essays. Briefs built on questions produce answers, and engines cite answers.
Read the current answer before writing a new one
Take the question and run it through the engines your buyers use. Read the full response, then look at who got cited and click through to see why. You are after two things: what the engines currently say, and what earned those particular sources their seat.
The pattern shows up fast. The cited source is rarely the biggest brand or the longest page. It is the source that stated something plainly and put evidence next to the claim. The plainest defensible sentence in the neighborhood of the question tends to win.
You will also see how weak the incumbent answers often are. Hedged, generic, stitched together from me-too vendor pages and a five-year-old forum thread. We map this at category scale inside an AI visibility audit, but a single brief does not need the full map. An hour of running prompts and reading citations will show you the opening.
Find the claim gap
The claim gap is the thing your piece exists to close: something true, useful, and squarely on the question that no citable source currently says plainly.
It is usually not a new idea. It is a missing plain statement. Nobody has said under what conditions the standard advice fails. Nobody has put the definition, the criteria, and the trade-offs on one page. Nobody has answered the question actually asked, only the adjacent easier one. Your subject matter experts say the missing thing on calls every week. It has simply never been published in a sentence a machine can lift.
One gap per piece. A brief that tries to close four gaps closes none.
The one-page brief
Ours has six lines.
- The one question this piece answers, in the buyer's exact phrasing.
- The extractable claim, stated in one sentence an engine could quote verbatim and still be right.
- The evidence that backs the claim, with the instruction to place it beside the claim, not three scrolls below.
- The entities that must be named consistently: product, category, people, spelled the way they are spelled everywhere else on the site.
- The internal links, chosen to reinforce the pages that matter, not sprayed.
- The skip conditions: what would make an engine pass us over.
The second line is the heart of the document. A brief that cannot state its extractable claim in one sentence is a brief for a piece that will not be cited. Engines quote sentences. If the person writing the brief cannot produce the sentence, the writer will not discover it by accident in paragraph fourteen, and the piece will read fine and get quoted never.
The last line earns its keep too. Ask what would make an engine skip us and the answers get usefully specific: hedging the claim into mush, burying it under six hundred words of throat clearing, contradicting our own pricing page, calling the product three different names in one piece. Write the failure modes into the brief and the draft avoids them without a fight.
What it looks like when it compounds
We ran briefs exactly like this for a Series D identity verification platform, one question, one claim, one gap at a time. They went from sixth to first in AI visibility in their category in a single quarter. The writers did not get more talented in ninety days. The briefs got more specific.
The quiet benefit is that none of this fights good writing. A piece that answers a real question with a plain claim and nearby evidence is also the piece a human buyer actually wants. Citation engineering, done at the brief stage, is mostly the discipline of deciding what you are willing to say, and that discipline is the working unit of AI search optimization: not the campaign, not the migration, the claim.
Next time a brief crosses your desk, ask it for its one sentence and see whether it has one.