Searches for “ai visibility” are up 400% year over year. “ai visibility audit” is up 1,200%. I pulled both numbers this morning, and the second one is the more interesting of the two, because a business owner who searches for an audit has already decided the thing is real and now wants to know where they stand.
Then I looked at what people search alongside it. In the last 30 days in the US, the top related queries for “ai visibility” are Profound, Scrunch, and Searchable. Those are all vendors. Nobody is searching to learn the concept. They are shopping for a tool that will hand them a number.
So here is the honest version of what that number is, what it is worth, and what it cannot tell you. I run this audit every week for clients and for anyone who asks for a scan, and there are exactly four numbers that matter.
The screenshot problem
Most AI visibility reporting I see from other agencies is a screenshot. Somebody types a question into ChatGPT, the client’s name appears, the screenshot goes in the deck, everyone claps.
That proves nothing. Ask the same question twice and you can get two different answers. Ask it from a different account, in a different city, on a different day, and the list of businesses changes again. A single screenshot is one roll of the dice presented as a season record.
A real audit runs the same buyer question many times, across the platforms that matter, and reports frequency and placement. That gives you the first two numbers.
Number 1: presence rate
Presence rate is the percentage of runs where your business appears in the answer at all. Out of 100 people asking an AI assistant that question, how many see your name.
Here is what that looks like on a live account. This is a medical courier operating in 30 metros across 8 states, one of ours:
| Buyer question | ChatGPT | Google AI Mode | Gemini |
|---|---|---|---|
| medical courier | 75.5% | 93.9% | 75.5% |
| stat medical courier | 87.8% | 83.7% | 70.8% |
On “stat medical courier,” which is the urgent, high-margin version of the query, they show up in nearly 9 out of 10 ChatGPT answers. On the broad term in Google’s AI Mode, 94%.
You can set a target against a number like that and check it again in 30 days.
Number 2: average rank inside the answer
Appearing is not the same as being first. AI answers are lists, and position inside the list still decides who gets the call.
Same account, same questions:
| Buyer question | ChatGPT | Google AI Mode | Gemini |
|---|---|---|---|
| medical courier | rank 2.35 | rank 1.39 | rank 1.62 |
| stat medical courier | rank 1.05 | rank 2.05 | rank 1.76 |
A 1.05 average rank on “stat medical courier” in ChatGPT means that when they appear, they are almost always named first. That is the closest thing to a number one ranking that exists in AI search right now, and it took months of Maps work, service-page structure, and consistent entity signals to get there.
Number 3: the platform split
This is the one that surprises people, and it is the reason a single-platform report is close to useless.
Look at the top row of that table again. Same business, same question, same week. Google AI Mode puts them in 93.9% of answers. ChatGPT puts them in 75.5%. That is an 18-point gap between two assistants on identical input.
The platforms read different sources. Google’s AI surfaces lean on the Maps and local corpus, which is why heavy Google Business Profile work pays off there first. This same client holds 67% share of local voice in Tampa, 69% in Clearwater, and 65% in Brandon on “medical courier,” and you can see that strength flow straight into the Google AI number. ChatGPT weights third-party mentions, directories, and text it can quote, which is a slower and more editorial build.
If your report only checks ChatGPT, you are reading one column of a three-column table and calling it the score.
Number 4: measured AI referral sessions
Everything above is what the assistants say about you. This number is what it produced.
A national property tax appeal firm we run, roughly 3,000 pages, recorded 146 AI referral sessions in July, counted in analytics rather than modeled. That is real traffic from people who asked an assistant a question and clicked through.
The same account is cited by ChatGPT on “best property tax appeal service,” named as one of only two sources in the answer. The other source is a competitor whose credibility comes from a MarketWatch article. We got there with comparison pages and county-level coverage instead of a national press placement, which is a route almost every mid-market operator can actually run.
For context on that account’s month: 7,836 clicks from Google, up 52%. Click-through rate 1.20%, up from 0.92%. Average position 7.8, improved from 8.2. Traffic grew while CTR and position both got better at the same time, which almost never happens, because growing sites normally dilute.
What AI visibility does not measure
Volume. And this is where most people misread their own report.
Take the courier account again. Only 22% of its search clicks are new discovery. The other 78% are people who already know the company. Out of roughly 389 monthly discovery clicks, one of them became a client now running close to $3,000 a week in accounts receivable, roughly $150,000 annualized, and that client pulled the business into two new markets that already had pages built and ranking before anybody needed them.
389 clicks is a rounding error to a consumer brand. One of them paid for the entire program many times over. When your buyer is a business signing a recurring contract, the presence rate on the four questions your buyer actually asks is worth more than a traffic chart with a nice slope.
I learned this on my own money before I ever charged anyone for it. At my elearning company I ranked 360 keywords in the top 1 to 3 across 156 product pages, and when I sold that company to private equity, the search asset was a material part of the valuation. Nobody in that diligence process asked about sessions. They asked whether the demand was durable and whether it belonged to the business.
How to run this on your own business
You can do a rough version yourself this week:
- Write down the four questions a buyer asks right before they hire someone like you. Use their wording, not your service names.
- Run each one 20 times in ChatGPT, in Google AI Mode, and in Gemini. Fresh session each time, no personalization.
- Record two things per run: did you appear, and where in the list.
- Check your analytics for referral sessions from AI assistant domains.
That gives you presence rate, average rank, the platform split, and measured referrals. Four numbers, one afternoon.
Then the harder question, which is what to fix. Presence gaps in Google’s AI surfaces usually trace back to Maps and location coverage, and I wrote about how one Google Business Profile can carry 30 metros in multi-location SEO. Gaps in ChatGPT usually trace back to what other sites say about you and whether your own pages answer questions in quotable form, which I covered in getting recommended by ChatGPT. The full picture of how these pieces fit together sits on our AI SEO agency page.
Want the four numbers without the afternoon
Run the manual version if you want to understand the mechanics. I recommend it, honestly, because doing it by hand is how you learn which questions your buyers actually ask.
If you would rather just see where you stand, we do this for free. The AI visibility scan runs your buyer questions across ChatGPT, Perplexity, and Google AI Overviews, reports your presence rate and placement, checks whether the AI crawlers can even reach your site, and comes back with the specific gaps to close first.
No pitch attached to it. You get the numbers either way.