You have read the first report and you know which questions get answered without you. This guide turns that gap into one page: who it is for, which question it answers, where its numbers come from, and which version goes live.
Find the empty cell first
After every collection the coverage analyst runs on its own and produces a table of audience by stage. For each audience it counts the questions at awareness, comparison and purchase, and how many of those answers named you. An empty cell is one with questions in it and no you.
For most brands the gap is at comparison. Awareness questions are broad and everyone gets a line. Once the question becomes “A or B”, an AI only names the brands it has read a comparison about.
Pick the questions where the competitor is named and you are not
Empty cells hold two kinds of question. In one, nobody gets named and the answer stays generic. In the other, a competitor is named and you are not. Write for the second kind first. The AI is already willing to give a specific name on that question; what you lack is a page it can read.
On the Brand tab, open “Topics & Questions”, regroup by stage, and expand comparison. Under each phrasing you can see who every platform named. Or ask the AI CMO directly: “which comparison questions name B and not us”.
Dispatch from the suggestion card, scope already filled
At the end of the chat turn after a collection there are at most three suggestion cards, one per stage. The comparison card is usually this job. Tap “send to the web writer”. The scope (which keyword, which phrasings, which audience) is filled in by code, so the writer writes for that one keyword and does not drift into an industry overview.
Every figure needs a source
The web writer writes long form with citations. Every number in the draft has to point at something: an asset you uploaded, data from a connection, a public report. A number without a source gets the draft sent back to the writer once. If the second pass still has no source, that number is flagged and put in front of you, and you decide whether to supply the source or cut the figure. The writer does not invent one.
The rule feels strict, and it is the reason the page can be cited at all. What a page that gets cited looks like is in A web page that gets cited.
Edit in Solve, every save is a version
Drafts live on the Solve tab. Edit them yourself, or say in chat “the intro is too long, cut it to three sentences” and the writer edits. Every save is a version with its author on it, yours or the writer’s. You can compare any two versions to see which sentences changed, and restore any earlier one.
Editing an existing draft is always a new version of the same draft, never a second file. Asking the writer for a rewrite is also a new version of the same draft, and so is the writer changing the title. The draft list will never contain “comparison v2 final”.
Publish, then collect again
Publishing to your website goes through GitHub Pages or your own server over SSH, with a backup taken before anything is replaced. After that there is nothing to do but wait for the next collection. Compare the three phrasings of that keyword across the two collections and see whether any answer now names you. If the cell in the coverage table changed, the page is done. If it did not, check whether your page appears among the cited sources; if it does not, the problem is indexing, not the writing.
Before you start
- Collect at least once. The coverage table needs data before there is an empty cell to find.
- Upload what you have to the assets section on the Brand tab. The writer looks there first for sources.
- Set up the website connection, or the finished draft has nowhere to go.
Further reading: Climb from the tier you are on · A web page that gets cited