What leadership is actually asking
Two things we heard directly in August 2026. A marketing lead at a large payments company: "I need to make my case clearer to higher-ups." A founder, on the same topic: "all the AEO tools are scams." Both are the same request from different sides. Show me something about my company, on questions my buyers ask, that I can check myself.
A brief that opens with a definition of generative engine optimisation fails that request. A brief that opens with a table of your buyers' questions and who gets named on them passes it, because every cell can be reproduced by the reader with a browser.
The one-page structure
| Section | Content | Length |
|---|---|---|
| 1. The table | Prompt, who is named, who is cited. Six to twelve rows, one engine, dated. | Half the page |
| 2. The reading | Two sentences: where we are known, where we are absent, who owns the jobs we are absent from. | Two sentences |
| 3. The one change | One URL, what it will say after the change, who owns it, ship date. | Three lines |
| 4. The re-run | The date the same prompts will be run again and the two numbers you will report then. | One line |
Section 1: the table
Pick the prompts sales hears on calls, in the buyer's words. Run each in ChatGPT with search on. Record who is named in the answer and which URLs are cited. Keep named and cited as separate columns; a link to your page with no mention of your brand is a different fact from a recommendation, and Semrush measured that gap at about 62% of all AI citations. Date the table.
| Buyer's question | Named | Cited | Run on |
|---|---|---|---|
| tool to see why ChatGPT recommends a competitor | Competitor A, Competitor B | competitor-a.com/guides/… | 1 Sep |
| how to measure brand mentions in ChatGPT | Competitor A, us | none | 1 Sep |
| what is [our brand] | us | ourbrand.com/ | 1 Sep |
Section 2: the reading
Two sentences, no adjectives. "We are named on brand questions and absent on the three buying questions. Competitor A is named and cited on all three, from one guide page each." That is the whole diagnosis, and it is the kind of sentence a CEO forwards.
Section 3: the one change
Name one URL. Say what its first paragraph will answer after the edit. Give it an owner and a date. Resist listing the roadmap. A brief with one change and a re-run date is credible; a brief with twelve initiatives is a budget request in disguise, and it will be read as one.
Section 4: the re-run
Commit to running the identical prompts again on a named date and reporting the same two numbers. This is what separates a measurement from a pitch. If the numbers do not move, say so; a null result on a single page change narrows the next test. The trial protocol on the research hub is the long form of this discipline.
What to leave out
- The category. Do not explain GEO, AEO or their difference on the first page. Link to a definition in a footnote if someone asks.
- A single blended score. "Our AI visibility is 34" invites the question "34 what", and the honest answer takes a page. Two counts on a table are self-explanatory.
- A promised lift. "We expect a 40% increase in mentions" is unfalsifiable and everyone in the room knows it. "We will re-run on the 15th" is a commitment.
- Tool names and screenshots of dashboards. The evidence is the table. The tool is how you keep producing it.
- Anything about the AI market, adoption curves or the future of search. They have read it. They want to know about their company.