We asked AI about 31 med spas and cosmetic dentists. 14 were never named.

Five buyer questions, three assistants, one run each. Nearly half the practices we checked did not appear in a single answer, and the sources behind the answers were mostly not the practices’ own sites.

Shaik Arif Hussain

Local SEO and Google Business Profile consultant since 2019. He has helped more than 650 business profiles through suspensions, reinstatement, verification and ownership disputes.

29 October 2026 · 7 min read

On 28 and 29 October 2026 we ran our free scan for a list of local businesses we were thinking of writing to: five buyer questions, each asked once of ChatGPT, Gemini and Perplexity, through each company’s own connection with its web search switched on. For every business we recorded how many of those answers named it.

This post is the count, for every trade in the list and then for the two medical trades in it: cosmetic dentists and medical spas. No business is named here, including the ones the assistants recommended instead.

It is a small sample, measured once. The limits are set out in full further down, and they matter more than any single number above them.

We checked 31 medical practices — 18 cosmetic dentists and 13 med spas — in Austin and the Denver area. 14 of the 31 were not named in a single answer. That is 8 of the 18 cosmetic dentists and 6 of the 13 med spas.

Across all eight trades we scanned, 149 scans came back readable, and 58 of them (39%) found the business named nowhere. The medical practices were a little worse off than the average, not a different story.

Nobody was named everywhere. Of the 108 businesses where all fifteen answers could be read, none was named in all fifteen; the best was named in 13. Among the 23 medical practices with all fifteen answers read, the best was named in 9.

For the 23 medical practices where all fifteen answers could be read:

The middle practice was named in none. Across all 31, including the scans where some answers could not be read, the practices were named in 64 of the 433 answers we read about them — about one in seven.

The same shape holds across every trade we scanned. Of the 108 complete scans, 48 were named in none of the fifteen answers, 30 in one or two, and only 6 in nine or more.

Charlotte is missing, and the reason is ours. All 21 medical practices we found there came back with no readable answers, because the account that pays for two of the three assistants, and for the model that reads the answers, ran out of credit that day. We would rather show a gap than fill it with a guess.

Across all trades, Austin had 85 readable scans and 37 businesses never named; the Denver area had 52 and 20. Charlotte had only 12 readable scans, too few to say anything about. Plastic surgeons were on the list too, but only two were found, in a test run, and neither was scanned, so there is no plastic surgery figure here.

The answers themselves are stored, so we could also count the sources each assistant showed. This covers the 87 medical answers from Austin, Denver and Greenwood Village, a Denver suburb where some practices are based.

The three assistants read very differently. ChatGPT showed a median of 5 sources per answer, Gemini 7, and Perplexity 32. Between them they cited 1,270 sources across 311 different websites.

The 31 practices’ own websites accounted for 114 of those 1,270 — about 9%. The rest were other practices’ sites and a familiar cast of third parties. Counting how many of the 87 answers cited each:

Two of those are worth a second look. Delta Dental, an insurer, was cited in 7 of the 9 answers to “dentist accepting new patients”: for that question, the insurer’s directory was the source more often than any practice. And no answer listed a Google Maps page among its sources. That does not mean the profile is unused — Gemini’s search runs on Google’s own results — only that it was never shown as a source.

Of the 17 practices named at least once, 14 also had their own website among the sources. Of the 14 never named, 4 did.

That is an association, not proof of cause. When an assistant names a practice it often links to that practice’s site, so the arrow can point either way. What it does show is that the practices getting named were, almost always, practices whose own pages were being read.

One of the five med spa questions was about money: “botox in” the city, “cost”. All nine answers to it quoted dollar figures. ChatGPT’s three answers cited four sources between them, every one a practice’s own page, and two of the four were pages about pricing.

Nine answers is a very small sample. But it is a clean illustration of something we see across trades: when somebody asks what a treatment costs, the assistant answers with somebody’s published numbers, and a practice that publishes none cannot be the one quoted. The practical version is in what a medical practice should publish.

The same five for every practice in a trade and town, with the town filled in:

• Cosmetic dentists: “can you recommend a good cosmetic dentist in [town] that people are happy with”, “i need a cosmetic dentist in [town], who should i actually call”, “who is the best cosmetic dentist in [town] and why would you pick them”, “dentist in [town] accepting new patients”, and “emergency dentist in [town] today”.
• Med spas: “best med spa in [town]”, “can you recommend a good medical spa in [town] that people are happy with”, “i need a medical spa in [town], who should i actually call”, “who is the best medical spa in [town] and why would you pick them”, and “botox in [town] cost”.

None of them names a business, because naming one manufactures the answer. Some are terse, the way people type into a search box; our question library was rewritten into fuller sentences on 29 October, after these scans ran.

It is one run. Each question was asked once of each assistant, and the same question gets a different answer twice. A practice named in 0 of 15 might be named in 2 of 15 tomorrow. Treat each practice’s number as a single draw and the table as the pattern.

It is small. 31 medical practices in two metro areas, two trades, two days. Nothing here is a national figure, and none of it is a ranking.

The list was not random. It holds businesses Google returned near the top of a search for the trade in that city, that have a website, and that publish an email address. Practices that are harder to find may do worse; we did not measure them.

Some answers could not be read. Fifteen answers were asked for each practice, but for 8 of the 31 our reader failed on some of them, so those practices were judged on between 8 and 14. The tables above say which figures use only complete scans.

It is a snapshot. Assistants change how they search and what they show, and this is late October 2026.

It is not the consumer app. We ask through each company’s official connection with web search on, not through the app on somebody’s phone, which personalises. How we ask, in full.

If you run a practice, the useful question is not the average. It is whether you are one of the fourteen. That takes a few minutes to find out, and the two guides that follow this post say what to look at first: what AI reads before it recommends a practice and what to publish so it can describe you. Nobody can promise you a place in an answer, including us. What can be measured is whether you are in one now.