Five numbers on one page: a fixed set of buyer prompts, the share of those prompts where your company is named, the share where a competitor is named instead, the share where you are cited without being named, and the sessions that arrived from AI engines and what they did. Everything else is decoration. Semrush’s 2026 AI Visibility Index, built on 126 million US prompts run between January and April, found that 45% of marketing leaders cannot accurately measure their brand’s presence in AI answers and only 9% have tools that track every metric they need. The CEO does not need a dashboard that proves the category is changing. The CEO needs to know whether the company is in the answer, who is there instead, and what one decision would change the next reading.
Why the first report is usually wrong
The first AI visibility report most companies produce is a screenshot. Someone asks ChatGPT “best providers of X” on a logged-in laptop, the company appears or does not, and the screenshot travels up the org chart as evidence. It is evidence of one answer, on one engine, on one afternoon, shaped by that laptop’s chat history. Semrush’s index found ChatGPT cites an average of 15 sources per response while Gemini cites 3, so the same prompt can produce a crowded answer on one engine and a two-name shortlist on another. A brand that is visible in one is routinely invisible in the next. A single screenshot cannot tell the CEO which of those worlds the company lives in, and a report built on screenshots tells the CEO whatever the person taking them wanted it to say.
The second common failure is reporting citations as if they were recommendations. In June 2026 Semrush and Kevin Indig ran 115 prompts across 14 countries on ChatGPT, Gemini, Google AI Overviews and AI Mode, producing 3,981 domain appearances, and found 62% of them were ghost citations: the page is linked as a source and the answer never says the brand’s name. The engines split in opposite directions. When a brand appears in Gemini it is named in the text 83.7% of the time and cited as a source 21.4% of the time; ChatGPT cites 87% of the time and names 20.7%. A report that counts links is measuring how often the company did the reading for the machine. A report that counts names is measuring whether the buyer heard of you. Those are different products, and only one of them sells anything.
The five numbers, and where each one comes from
The first number is the prompt set itself, and it is a number because its size and stability are what make everything after it comparable. Thirty to fifty questions a real buyer asks before they contact anyone: the category shortlist, the comparison against your two nearest competitors, the “is X worth it” question, the pricing question. Fixed in writing, run monthly on the same three or four engines, changed only by decision and never by convenience. Without this, every subsequent report measures a different thing and the trend line is fiction.
The second number is share of mentions: on what percentage of those prompts is the company named in the answer text. The third is the same number for each named competitor, on the same prompts, which is the line the CEO will actually look at. The fourth is the ghost rate, the share of appearances where a page of yours was cited but the name was absent, because that gap is the cheapest thing to fix and the study found comparative content produces 2.4 times more brand mentions than informational content. The fifth is what the engines send you. Google Search Console’s generative AI performance report, launched on 3 June 2026 and rolled out to every site worldwide on 31 August, reports impressions in AI Overviews, AI Mode and generative features in Discover, by page, country, device and date, with no clicks, no click-through rate and no queries. Bing Webmaster Tools has reported AI citations since 9 February 2026 and added Citation Share on 16 June: your citations for a query divided by all citations shown for that query. Both are free, both are first-party, and neither existed a year ago.
Sessions from AI engines belong on the page too, with a caveat attached. Adobe Digital Insights reported in June 2026 that AI-referred traffic to US retail sites grew 138% year on year in May, is up 1,324% since October 2024, and converted 54% better than non-AI sources. Retail is not your business, but the shape is: the volume is still small, the intent is high, and the growth rate will make anyone’s chart look impressive for the next two years regardless of whether anyone did anything. Report the sessions and the conversions. Do not report the growth rate as an achievement. It is the weather.
A report that counts links measures how often the company did the reading for the machine. A report that counts names measures whether the buyer heard of you.
What you can actually do
The report is a CEO’s instrument, so the CEO should specify it. The Signal has covered how to measure brand in AI search and how to track AI search traffic; this is the version that fits on one page and survives a board meeting.
Write the prompt set yourself
Thirty to fifty questions, in the words a buyer would use, signed off by the CEO and the head of sales. Ask sales for the five questions prospects say they asked an assistant before the first call. The set is the baseline. It does not change because the results are embarrassing.
Demand names, not links
The headline metric is share of prompts where the company is named, next to the same figure for two competitors. Citations go on the page as the ghost rate, labelled as a repair list rather than a win.
Turn on the two free reports this week
Search Console’s generative AI report and Bing Webmaster Tools’ Citation Share cost nothing and take an afternoon. Any agency or tool that reports AI visibility without them is reporting its own sampling instead of the engines’ records.
One trend, one risk, one decision
The report ends with the one prompt where a competitor gained on you, the one page that is cited but never named, and the one thing the company will publish or fix before the next reading. If it ends with a chart instead, it was for the marketing team, not the CEO.
The part nobody wants to hear
The report is easy. The prompt set takes an afternoon, the two free reports take another, and a competent marketer can produce the five numbers by the end of the month. What most companies cannot produce is the decision at the bottom of the page, because the decision is nearly always “publish something specific that we have avoided saying,” and that requires a person with the standing to say it. Peec AI’s March 2026 analysis of 30 million citations found that even the most cited domain on any engine holds under 5% of the total, which means nobody owns the answer and the answer goes to whoever is clearest. Clarity is a leadership output. The CMO Survey found four in ten companies already running some form of generative engine optimisation, and the same survey found no marketing technology capability scoring above 5 on a 7-point scale, two years running. The tools are in. The operating system that turns a monthly reading into a monthly decision is, in most mid-market companies, a person with ten spare hours a week, which is a person who does not exist.
So the first report gets made, it goes to the CEO, it says the company is named on 14% of buyer prompts and a competitor on 31%, and then it is the second report that tells you what kind of company you are. If the number moved, someone owned it. If it did not, the report was measured and never acted on, which is the most expensive kind of visibility. Be the answer, not the ad, has always meant deciding what the company is willing to be the answer to, and then being it, on a schedule, in front of a machine that keeps score. What gets measured gets cited. What gets decided gets named.
The first AI visibility report should be five numbers around a fixed prompt set, headlined by how often the company is named against its competitors, and it only matters if the CEO can point to the decision the second report will test.
A first report like this is what the AI Opportunity Audit produces in its opening week, before any recommendation is made. The AI consulting pages for Connecticut, Massachusetts and Rhode Island describe the rest of the engagement.
One check belongs in that first report: how much of the traffic being counted is automated. Is bot traffic making our marketing numbers wrong? covers it.