How The Named List counts a name, not a score
The Named List counts how often a business is named in answers to a set of customer questions, on a stated run, with the assistant and the date recorded. The result is written as named in X of Y questions. It is not a score out of 100, and it is not an official "AI visibility score" from OpenAI, Google or anyone else. A name in one run can be missing in the next. AI results vary.
The same rule is on the method page. This page exists so the counting rule is easy to find, and so it is not mistaken for a composite metric. The AI visibility checklist is the do-it-yourself list that uses this rule at the end.
What counts as a name?
A business counts when the answer text names it. The method page puts it this way: a business whose website was read, but not recommended, does not count as named.
That distinction matters:
- Named: the answer text includes the business's name as one of the places to use, call or visit.
- Cited: the answer links to, or points at, a source. That source might be your site, a directory, or someone else's page. Being cited is not the same as being named. The difference is explained in named or cited? What ChatGPT results actually mean.
- Found but not named: the assistant retrieved your page, then recommended other businesses. In the published physiotherapy sample, that happened for one further question besides the one name. In the dental checks, it happened in 6 of 52 answers. Those figures are on the sample report and the dentists page.
If you are scoring a test at home, write three columns, not one: named, cited, absent. Do not turn those marks into a percentage and call it a score.
How is a run recorded?
Every published count on this site is tied to a specific test, not to a rolling average. Record at least:
- The exact questions, with the town or area in each one.
- How many questions you asked (Y).
- How many of those answers named the business (X).
- Which assistant you used, and whether live web search was on.
- The date, and if useful the time and the location setting.
- That it was one run, unless you are explicitly repeating the same questions on the same day, in which case record each run separately.
Checks so far use Claude with live web search. I also put the same kind of questions to ChatGPT, used logged out with no account, and note which businesses it names. The report names the assistants used and the date. More assistants can be added to a check. Do not treat a Claude count as a forecast of ChatGPT, or the other way around.
Write the questions the way a customer asks who to call, for the work you actually do. Do not test only "what do you know about [your business name]". That tells you whether the assistant can find you, not whether you are named when someone asks who to use. A way to run the test yourself is in how to check if ChatGPT recommends your business.
What do the published examples show?
These figures are already on the site. They are snapshots, not promises, and they do not name the businesses.
On the sample report, a private physiotherapy clinic in a UK city was tested with Claude, using live web search, location set to the clinic's city, on 10 patient-style questions, in a single run. It was named in 1 of 10 answers, even though it had a strong Google rating. Two competitors were each named in 6 of 10, and two more in 5 of 10. For one further question the assistant read the clinic website, then recommended other clinics.
Also on that page, a dental practice in Birmingham was tested with Claude, with web search on, on 16 October 2026, location set to Birmingham, on 4 customer questions, in one run. It was named in 1 of 4 answers. Another practice was named in all 4.
On the dentists page, I asked the same four patient questions for 13 private practices in Greater Manchester and the West Midlands, on 16 October 2026, with Claude and live web search, location set to each practice's city. The practices were named in 21 of the 52 answers. Three were not named in any. Ten of the 13 were named at least once. One was named in all four answers. In 6 answers, Claude's search returned the practice's own website, and the answer still recommended other practices. Claude pulled 608 pages to write the 52 answers. 46 answers drew on at least one dental directory, review or listing site. 29 answers quoted a star rating or a review count. Three Manchester questions were run three times on the same day, from the same location, and the practices named changed between runs for all three.
One run per question unless stated otherwise. AI results vary between runs, apps and users.
Why not a score out of 100?
A 0-100 "AI visibility score" has to mix things that are not the same: different assistants, different questions, different days, citations versus names, and sometimes traffic or listing completeness. There is no official scale for that mix. OpenAI's Business Hub Insights, where they are available, count mentions in a specific way for online and local businesses, over 28 days, and a blank metric does not necessarily mean zero activity. Google's Search Console Generative AI performance report measures impressions in Google's own AI features, not ChatGPT. Neither is a score I would turn into a single number for a local business.
A dated count says what happened when these questions were asked, on this assistant, on this day. It can be compared with a later re-test of the same questions. It cannot promise that the next customer will see the same names.
If a third-party tool shows you an "AI visibility score", ask what it is actually counting, on which assistants, and whether a citation is being treated as a recommendation. If the answers are vague, treat the number as marketing, not as a measurement.
Frequently asked questions
Is this an official ChatGPT or Google score?
No. OpenAI and Google do not issue a public 0-100 AI visibility score for a local business. This is a dated count of names in answers to stated questions.
Does a citation count as a name?
No. A business counts only when the answer text names it. A link, a map pin or a page the assistant read is a citation or a source, not a name.
Why do the published checks use Claude if the site is about ChatGPT?
Checks so far use Claude with live web search. I also put the same kind of questions to ChatGPT, logged out, and note who it names. The report always says which assistants were used and the date. AI results vary between assistants.
If I am named in 1 of 10, is that a fail?
It is a count, not a pass or fail. It tells you what happened on that run. The next run can be different. The published physiotherapy example was named in 1 of 10, and still produced useful work: the assistant only treated the clinic as a sports injury practice.
Want someone to run the checks for you?
A Who-Gets-Named Check uses this counting rule. The steps after a check are on the method page. If you would rather hire someone, how to choose an AI visibility service lists the questions to ask, including of me. AI results vary.