Somebody three miles from your shop just typed "best physio near me" into ChatGPT. They got a short list. Three businesses, maybe four, written as a recommendation rather than a page of links. There was no second page to scroll to, no ad slot to buy, and no report that told you whether your name was on it.
That is the part worth sitting with. Search used to fail loudly: you could open a rank tracker and watch yourself sit at position eleven. An AI answer fails silently. You are either named or you are not, and nothing about your day changes either way.
Ten blue links became one short answer
The economics of a page of results and the economics of a recommendation are not the same thing. On a results page, position eleven still gets clicks from the stubborn. In an answer, position eleven does not exist — there is a first name, a second, a third, and then the answer ends.
That compression is the whole story. Every question becomes close to winner-take-most, and the businesses that get named collect a growing share of the intent that used to spread across a page.
The answer is not one answer
Here is the thing rank tracking taught everybody the wrong lesson about: an AI answer is not a single fact about your business. It changes with where the person is standing.
Ask from one side of town and you get one short list. Ask from the other side of the same town and you get a different one. Ask a slightly different question and the list reshuffles again. A single check tells you what one imaginary person saw once, which is a fact about that check rather than about your visibility.
This is why we scan on a grid rather than sampling one point. A grid of points around your location, the same question at each, and the answers stored — that is a map of where you are named and where somebody else is. One sample is an anecdote. A grid is a measurement.
Five surfaces, not one
"AI search" is not a single product either. Our AI Visibility view tracks five answer surfaces — ChatGPT, Gemini, Google AI Overviews, Google's AI Mode and Grok — because they disagree with each other constantly. A business that is the obvious answer on one can be invisible on another, and the reasons are not interchangeable.
What you get out of tracking all five, on a schedule, is the two things a single check can never give you:
- Trend. Whether last month's answer is this month's answer, per surface.
- Share of voice. Who the answers recommend instead of you, and how often. Competitors are not something you list in a settings page here; they are whoever the answers actually name.
What to do with the map
The honest answer is that nobody can promise you a ranking in a generative answer, and you should be suspicious of anybody who does. What a map gives you is direction, and direction is most of what strategy is:
- Find the holes. Where are you named, where are you not, and is the boundary geographic or does it track the question being asked?
- Check who is in the answer instead. Competitors in an AI answer are often not the competitors you argue with about price. Read what the answer says about them — it is usually quoting something.
- Fix the sources. Answers are assembled out of what the web says about you: your own pages, your profiles, the places that list you. That is ordinary work, and it is still the work.
- Measure again on a schedule. Not once. The interesting number is the delta.
Start with one free scan
The reason we put an AI visibility checker at /tools/ai-visibility-checker with no account and no email wall is that the first look should cost nothing. It runs a real grid scan of one keyword on one surface and shows you what comes back — including which competitors get named in your place.
If it turns out you are the answer everywhere you sell, that is a genuinely good morning and the tool cost you nothing. If it turns out you are not, you have found the thing that was quietly happening without you.
The businesses that get named in 2027 will be the ones that started looking in 2026.