How to Report AI Visibility to a Client
Reporting AI visibility is harder than reporting rankings, for three reasons that all arrive at once. The metric is unfamiliar, the output is noisy, and the first months frequently show no movement at all.
Handled badly, that combination ends the engagement in month three. Handled well, it produces a client who understands why the work takes time — which is a considerably better client.
Set the expectation before the first report
The conversation that prevents most problems happens before any data exists.
Say plainly: this moves slowly, the first six weeks will probably show nothing, and the number will bounce month to month regardless of what we do. Then explain why — platforms re-crawl on their own schedule, and the outputs are probabilistic by design.
A client told this in month zero reads a flat month two as expected. A client told it in month two hears an excuse.
Freeze the question set, and show it
The single most important methodological decision is that the questions never change between runs. Change them and the trend line is not a trend line.
Show the client the actual questions. Not a summary — the list. Two things follow. They can tell you which ones their buyers would never ask, which improves the set. And when the number moves, they know exactly what moved, because they have seen the input.
A visibility score from a question set the client has never read is a number they have no reason to trust.
Report four things, in this order
1. Presence, split by intent
Not one number. Four: brand, category, problem, comparison.
The aggregate flatters everyone, because brand questions are easy and inflate it. Split by intent, the story is usually clearer and more honest — strong where people already know the name, weak where new buyers start.
2. Who appeared instead
The most-read part of any report I have seen. A ranked list of every brand named across the question set, with counts.
Clients care about this more than their own number, and it routinely surfaces competitors they had not considered. It also makes the work concrete: these are the companies currently occupying the answers.
3. Which sources were cited
The domains appearing beneath the answers. This is where the report becomes a plan rather than a scoreboard — it converts "we are not visible" into "we are absent from these six pages".
4. Readiness, separately
Keep technical readiness apart from visibility. It moves on a different clock, improves fast, and gives you something honest to show in the early months when visibility has not moved.
Just do not present it as visibility. A readiness score climbing while citations stay flat is progress on inputs, and saying so plainly is what keeps you credible when the outputs eventually move.
What not to put in the report
Screenshots as evidence. One answer naming the client proves nothing — ask again in five minutes and it may not. Screenshots are useful as illustration and dangerous as proof, because the client will re-run it and get something different.
Traffic attributed to AI mentions. When a model names a brand without a link, no referral is recorded. Any number claiming to measure that traffic is modelled, and presenting a model as a measurement is how trust ends.
Causation you cannot support. The score rose after you published three pages. It also rose after a platform update and ordinary variance. Say what you did and what changed; resist joining them with a straight line.
The monthly rhythm that works
- Same questions, same platforms, same week of the month. Consistency is the whole methodology.
- One page of numbers, one page of what it means. The second page is what gets read.
- Three actions for next month, drawn from the cited-sources list rather than from general advice.
- A running log of what was shipped, so that in month six you can look back at inputs and outputs together.
The honest pitch
The most defensible position with a client is also the most accurate one: we cannot promise you will be recommended, because no one controls what these systems say. What we can do is remove every reason they cannot find you, put you on the sources they read, and measure the result the same way every month so you can see whether it is working.
Clients accept that framing far more readily than agencies expect. What ends engagements is not slow progress — it is progress that was promised faster.
