A good AI search report is short and mostly a table. Here is what belongs in it.
The six things
1. The prompt list, with two columns. Every tracked question, and for each: were you named in the answer, and was your domain cited as a source. Separately, not blended.
This is the whole report, really. The other five items explain it.
2. Who was named instead. Per prompt. This is the most actionable line in any AI search report and it is missing from most of them. Knowing that a specific competitor takes four of your ten prompts tells you what to do; knowing your score is 12% does not.
3. What was claimed or corrected. Which citation sources, which are still outstanding. Verifiable in ten minutes by looking at the listings yourself, which is exactly why it belongs in writing.
4. What was published, and which prompt it targets. One line each. A piece that targets no prompt should have a reason.
5. What was placed or pitched externally. The workstream most often absent from reports, because it is slow and produces nothing to show in a bad month. Its absence is itself informative.
6. What is next, and why that rather than something else. Two or three lines. Prioritisation is most of the value an agency adds, and a report that never explains a choice is hiding the interesting part.
Why two columns instead of one
Being cited as a source and being named in the answer have different causes and need opposite work.
A business in this position is being read constantly and recommended almost never. A single blended visibility figure shows a middling number and points at the wrong fix, which in that situation would be to publish more.
Any report that cannot separate these cannot tell you which problem you have.
Two metrics that mean nothing alone
Traffic. AI search often produces no click at all. Traffic can be flat while the work is going well, and can rise for reasons unrelated to any of it.
Volume published. An input. Twenty pieces that produce no lift is worse than four that do, because it also costs you the review time.
Neither is useless as context. Both are useless as the headline.
How to verify a report yourself
Twenty minutes, no tools.
Take five prompts from the report. Open a logged-out or temporary chat, because in your normal account the assistant has seen you discuss your own business and will bring it up. Ask each one. Record named and cited.
Your result should broadly agree with the report, allowing for genuine day-to-day variance in these systems. If the report shows movement you cannot reproduce at all, ask how the measurement runs: which engines, how often, logged in or out, and how a mention is counted.
What a report should not be
Long. A twenty-page deck for a monthly AI search engagement is a document rather than a report, and its length is usually inversely related to how much moved.
The six items above fit on two pages. If a report is longer than that and still does not contain them, that is the finding.
The frequency question
Monthly is right for reporting. Fortnightly measurement is right for the underlying data, because weekly checking shows you variance rather than progress and is a reliable way to talk yourself out of an approach that is working.
What one row of a good report looks like
Abstract descriptions of reporting are less useful than an example, so here is a single row of the kind of report worth paying for.
Question: "best general contractor for a full gut renovation in Austin" Engine: Google AI Overviews Checked: weekly, logged out Named: no last month, yes this month Cited: yes, both months Named instead: two competitors, one of them new this month Why it changed: state contractor license record corrected on the 3rd Next: same correction applied to two remaining license records
Everything in that row is checkable by the client in about two minutes, which is the point. A report you cannot verify is a story.
Note what is not in it: traffic, rankings, impressions, or a blended visibility score. Those may appear elsewhere in a monthly pack and none of them belong in the row that answers "are we being recommended yet".
A report of twenty rows like that tells you more than forty slides, and it takes less time to read.
What the reporting has shown
The figures above come from the same measurement that produces those rows, run continuously across every account rather than sampled monthly.
That is what makes a weekly view possible at all: 82,000 measured answers is not a number you can reach by checking a handful of prompts each month.
A worked example of reading a report
A month three report on a twenty-two question set, and what a client should take from it.
Named: 4, up from 1. Cited: 9, up from 3. Both moving, citations faster. That is the normal and healthy order, because citation sources are re-read frequently and being named lags behind them.
Two competitors named on six questions where the client is not. This is the most useful line in the whole report and the one most often missing. It converts "we are not there" into "these two are, on these six questions", which is actionable.
Claiming: three registers corrected, two review profiles claimed. Spot-checkable in ten minutes, and worth spot-checking.
Published: three pages, each mapped to a tracked question. The mapping is the part that matters. Three pages with no question attached is content, not AEO.
External: one pitch sent, one accepted for next month. Thin, and it should be. Month three is early for this workstream, and it is the one that eventually moves the named figure.
Next month, and why: the remaining two registers. A reason attached to a plan. A report that lists next steps without saying why those rather than something else is a task list, not a report.
See which questions name you
A free AI visibility audit on your own buying questions, across all five answer engines.
Book your audit
Ashur Homa
Built and scaled a digital brand to $100M+ in sales with zero ad spend. Has helped businesses generate millions through AI go-to-market strategy. Leads growth at Omni Eclipse.
Connect on LinkedIn
