A client forwards a screenshot. They asked an AI assistant for the best accountant in their city and it named three firms, none of them theirs. The email says "why aren't we in there, and what are you doing about it?" You can't answer with another screenshot. AI visibility can be reported honestly only if you separate what is measured from what is sampled or assumed, and a report that draws that line earns more trust than one with bigger numbers.
What can be measured today
Google's AI surfaces are measurable because they sit on a results page you can read. For a given keyword in a given market, you can record whether an AI Overview appeared, whether AI Mode answered the query, and in either case whether the client's domain was among the cited sources. Google's Search Central page on AI features and your website says there are no additional requirements beyond its existing Search Essentials to appear in them, which is why the same discipline as rank tracking applies: fixed keywords, fixed market, repeated on a schedule.
Two facts get tracked separately because they are different. Appearing means an AI answer exists for that query at all. Cited means the client's content is one of its sources. An AI Overview that exists and cites only competitors is the row worth acting on. A keyword with no AI Overview is a genuine zero, common for transactional and local queries, and it means "no AI surface exists here", not "the client is losing".
Mentions across the indexed web are also measurable: how often the brand's name appears in content Google has indexed, split by tone. The catch is the phrase. A count of mentions is a count of whatever phrase was searched, so a firm named after a common word gets the common word's count. A mention number in a report carries the exact searched phrase beside it, or it carries nothing.
What can't be measured, and how to say so
Whether ChatGPT, Gemini or Perplexity mention the client is a different kind of question. Those assistants generate a fresh answer every time, there is no results page to read, and no public ranking feed exists. Tools that show per-assistant mention counts either buy a data feed or show sample data, and some show sample data labelled as such while the feed is switched off.
The honesty rule follows. Sample data is a layout preview, not a measurement, and it never appears in a client report as a number. If a tile in your reporting tool is labelled sample, say so in the report or leave the tile out. A client who quotes a sample figure back to you has been told something that isn't true, and you're the one who sent it. The measured Google AI numbers usually make the same strategic point, and those you can defend.
For the assistants you can't measure, you can still observe. A fixed list of questions, asked the same way each month, with the named sources written down, is a structured observation. Report it as one, in words: how many fixed questions were asked, in how many the client was named, and last month's figure beside it. Never as a visibility score. Measuring your AI visibility sets that baseline up.
The AI page of the report
The AI section of a client report needs three numbers, each with its definition in the same line, shown as a trend rather than a snapshot.
- How many of the tracked keywords had an AI Overview or AI Mode answer in the client's market this month. This is the size of the AI surface for the client's keyword set, and it can be small for a local or transactional business without anything being wrong.
- Of those, how many cited the client, with the cited URLs listed. The gap between this number and the first one is the work.
- Brand mentions on the indexed web, with the searched phrase printed beside the count and the direction since last month.
AI surfaces change week to week, so a single probe is a snapshot. Report the direction over several checks and say how many checks the trend rests on. A tracker that records presence and citation per keyword over time gives you that history; a one-off audit probe gives you the baseline to start it from.
Explain what changed without taking the credit or the blame
AI surfaces appear and disappear for reasons unrelated to the client's site. A keyword that had no AI Overview last month and has one citing a competitor this month hasn't been "lost": a surface was created and the client wasn't in it. A citation that vanishes may mean Google stopped showing the Overview for that query at all, which the first number will show.
So the sentence for the AI page has the same shape as the one for rankings: what moved, what you did, and how sure you are they're linked. For a fictional dental practice: "AI Overviews now appear for most of the tracked treatment keywords, up on last month. The practice is cited in the implant answers, where we rewrote the page to answer the question directly, and not yet in the whitening answers, which we're working on now." Nothing there claims a cause it can't show.
What you're doing about it
The levers for being cited are the ones that make a page the clearest source for a question. The page answers the question directly, in a passage that can be lifted whole. It says who wrote it and why they know. Structured data describes what the page really is. AI crawlers aren't blocked by accident. And the brand is mentioned elsewhere on the web, because assistants talk about businesses the web talks about. Writing for AI citations and llms.txt and AI crawlers cover the page-level and crawler-level work. The report's job is to name which lever you pulled for which keyword, so next month's change has a candidate explanation.
What to take away
- Google's AI Overview and AI Mode are measurable per keyword and market; appearing and being cited are different facts and are reported separately.
- ChatGPT, Gemini and Perplexity mentions have no results page to read, so a number for them is a bought feed, a labelled sample or a manual observation, and the report says which.
- The AI page carries three defined numbers as a trend: keywords with an AI surface, keywords where the client is cited, and brand mentions with the searched phrase.
- AI surfaces come and go on their own, so the report names what you changed and how confident you are, and never claims a citation it can't connect to the work.
Next
With reporting honest across rankings, money and AI, the next chapter turns it into a meeting. The monthly review cadence gives you the structure.