GEO, case study · Nov 2026 · 13 min read

Why ChatGPT recommends other dental practices and not yours

Andrei Nekrasov

Most dentists have already run the obvious test: open ChatGPT, ask it for the best dentist in your town, and look for your own name in the answer. A few find it. The rest find 3 competitors, no explanation, and no one at ChatGPT to ask. Run the same question again an hour later and the list has quietly changed.

The natural conclusion is that AI works like Google’s map pack, only stricter. Keep the rating as close to 5.0 as you can, collect reviews, keep the website modern, and your name will come up. Every dental marketing company sells some version of that. It sounds right, because for 10 years that is close to how the 3 practices above the map got picked.

I spent October testing that on real practices. The rating carries no weight at all. A practice at 5.0 stars is as likely to be left out of the answer as a practice at 4.4. The groups AI names hold the same ratings as the groups it skips. 3 duller things separate them: how many people have bothered to leave a review at all, whether the practice has a page about the town where it actually sits, and whether its Google profile and its website tell a patient the same story.

Section 1

Google puts an AI answer above the links on 59% of the searches a dental patient runs

A patient searching for a dentist usually meets an AI answer before any link. That answer carries 3 practice names in it.

I ran 1,344 dental searches on Google, 56 in each of 24 US metros, and recorded whether Google printed an AI answer above the organic links. Google printed one on 795 of those searches, 59% of them. Hickory came lowest at 26 searches out of 56, Stockton highest at 40. So the AI answer reached every metro without taking over any of them.

Published measurements point the same way. 5W Research found a Google AI Overview on 49 of 59 dental searches. Conductor reports an AI Overview on 49% of page-one Google queries in health care. Patients also go straight to the chatbots: a BrightLocal survey of 1,002 US consumers found 45% had asked an AI tool to recommend a local business in the past year, against 6% a year earlier. AI tools now sit third as a source of local business recommendations, behind Google and Facebook.

Section 2

ChatGPT answers questions about dentists out of dental practice websites, not out of directories

In dentistry, ChatGPT reaches first for the practice's own website. The directories a practice pays for barely register.

Halcy ran 1,405 patient questions about dentists through ChatGPT across 20 US metros and classified all 13,152 sources the model cited, in work published by Dental Economics. Websites owned by dental practices were 54% of every source ChatGPT cited. National directories were 6% of them, review aggregators 4%, the magazine lists of top dentists 3%, and the websites of the large dental groups 0.7%.

Dentistry does not behave like the rest of health care here. Yext, across 6.8 million citations, found listings supplying 53% of what AI cited in health care, led by WebMD and Vitals. BrightLocal, across 1.9 million citations from local searches in every category, found business websites at 42% of all citations. So dentistry follows the local-business pattern, where the business website leads. The hospital and physician pattern, where directories lead, does not apply here.

My own searches show the same thing one city at a time. ChatGPT opened 4 local practice websites 5 times each while answering dental questions about Hartford. Gemini opened almost nothing except practice websites.

Section 3

Google's AI answer names the practices from the map pack and quotes their Google profile, not their website

Google and ChatGPT read different files about the same practice. A practice can therefore sit high in one while being absent from the other.

I asked Google for the best dentist in Charlotte, North Carolina. The AI answer named 3 practices and attached 3 sources to them. All 3 sources were Google Business Profile pages, with no practice website and no directory among them. The 3 practices Google named were the same 3 filling the map pack above the answer.

The citation counts from my own searches say the same thing. Across all the dental questions I ran, the engines opened google.com 1,779 times. The next domain after it, a national chain's website, came up 177 times. According to BrightLocal, Google Business Profile supplies 29% of everything AI cites in local answers, and Yelp 10%.

Section 4

31% of multi-location dental groups are absent from every AI answer

Most metros have already settled into one practice holding half the ground and several holding none of it.

I put patient questions about 24 US metros to ChatGPT, Gemini and Google AI Overviews, and counted which of 108 local dental groups came back. Every group in the count runs 2 or more offices. 34 of the 108 never came up at all.

The groups AI does name are not evenly matched either. In the median metro the most visible practice takes 51% of every mention the metro's practices got between them. 10 of the 108 groups take 42% of all mentions across the whole set. 11 groups came up on 10 or more questions out of 28. Another 30 came up on only 1 or 2.

SOCi found the same shape outside dentistry. SOCi measured roughly 350,000 locations across 2,751 multi-location brands and saw ChatGPT recommending 1.2% of those locations, Gemini 11% and Perplexity 7%, while the same brands reached Google's local 3-pack 36% of the time.

Section 5

Your Google review count predicts whether AI names your practice, and your star rating does not

A dental practice can see 2 numbers about itself in Google. One of them tracks being named in AI answers. The other tracks nothing.

The number of Google reviews a group holds tracks the share of answers naming it at 0.56, on a scale where 1 is a perfect match and 0 is none. The star rating tracks it at 0.02, which is no relationship at all. Sorting the groups into 4 equal bands by review count makes the size of the effect plain: in the band whose median is 28 reviews, 62% of groups were never named; at 372 reviews 35%; at 944 reviews 12%; at 2,670 reviews 7%.

Sorting the same 105 groups by star rating instead produces no gradient at all. 47% of the groups rated below 4.5 stars went unnamed, 17% of those between 4.5 and 4.79, and 27% of those at 4.8 and above, so the best-rated band holds more unnamed practices than the middle one.

Two independent studies land in the same place. 8ight.AI queried ChatGPT about 2,976 local businesses in 49 cities and found it naming 3% of the businesses holding under 25 Google reviews and 22% of those holding 1,000 or more. Spearleaf, whose 12 service categories include dentists, compared the businesses ChatGPT kept between 2 waves against the ones it dropped: the survivors held a median of 226 reviews and the dropped ones 164, while both carried the same 4.8 to 4.9 stars.

Practice size does not explain the relationship away. It still runs at 0.55 among my groups holding exactly 2 offices. The bottom half by review count went unnamed 41% of the time inside a single metro, against 16% for the top half. Review count does carry some information about how old and how large a practice is. The numbers therefore show a relationship, not a proven cause.

Section 6

Pages about the towns you serve earn AI mentions and pages about prices earn none

Some patient questions bring back practice names. Others bring back none. The patient's wording changes the answer.

My 28 questions per metro fall into 4 shapes. A question naming a suburb produced a practice name in 31% of answers, against 16% for a short query naming the whole city, 14% for a question written as a full sentence, and 13% for a plain request to recommend a dentist. The second wave repeated the order: 30%, 15%, 13% and 12%.

Address and distance drive that gap. Spearleaf found that holding an address inside the city the patient named raised the odds of being shown by a factor of 14.4, the strongest of every signal it tested. BrightLocal found 53% to 59% of AI recommendations sitting within 5 km of the searcher.

Price questions do the opposite. When a patient asks what a treatment costs, the engines answer with a price and name no clinic at all. I checked 10 price questions and then removed the whole question shape from the measurement, because there were no practice names in those answers to count.

Section 7

One check in ChatGPT is not a measurement, because the list changes between runs

The list a practice owner sees in ChatGPT today is partly a coin toss, so finding out where a practice really stands takes many questions asked several times over.

BrightLocal repeated 200,085 local searches and found only 23% of ChatGPT's recommendations, 21% of Google AI Mode's and 33% of Google AI Overview's holding from one run to the next. Spearleaf asked its markets again a week later and found the same business in first place in 34% of markets, with 48% of all the businesses it had shown appearing again at all.

I ran my own questions twice, about a week apart. The big picture held. For 84% of the 108 groups, the answer to "does AI name this practice at all" came back the same both times. All the movement sat at the bottom of the list. 17 groups changed sides between the 2 rounds. 16 of those 17 had either 0 mentions or 1. A practice named once this week can easily be named zero times next week. A practice on zero can gain a single mention without anything at the practice changing.

The same instability appears within minutes. Loading my Charlotte query a second time in one session returned a different third practice, both in the map pack and in the answer above it.

Section 8

Google's AI answer tells patients things a dental practice never published, usually about insurance

Vague wording on a practice website does not make AI careful, it makes AI fill the gap itself.

Herringbone checked AI answers about 600 small businesses, 200 of them dental and orthodontic practices, against what each business actually publishes. 27% of the AI answers on Google and 20% of the answers from ChatGPT carried information the business publishes nowhere: wrong hours, wrong insurance, treatments the practice does not perform.

Where a website left a detail vague, Google filled in a specific answer 80% of the time and attributed it to nobody, while ChatGPT did so 10% of the time and usually named the directory that supplied the detail. Herringbone's dental example: a question inside an orthodontic practice's blog turned into a claim that the practice accepts insurance, with carrier names added that the practice had never listed. SOCi, measuring separately, put the accuracy of business details on ChatGPT at 68%.

Section 9

Ask AI for the best dentist in your city and it names the national chains

The broader the question, the less room the answer leaves for a practice with 2 offices.

5W Research put 24 dental questions to 4 AI engines, 3 runs each, and counted the brands that came back. Aspen Dental appeared in 32% of the 288 answers, PDS Health in 19%, Zocdoc in 17%, Heartland Dental in 16% and Western Dental in 15%.

Halcy's data shows dental practice websites supplying 76% of the sources on orthodontic questions and 73% on periodontal questions, falling to 46% on general dentistry, where national directories climb to 15%.

Section 10

What to fix first, and how to check that it worked

Reviews come first because review count is the only thing that tracked being named in all 3 measurements on this page. The Google profile comes last because it only matters for the searches that run through Google.

Closing thought

So return to the test you ran at the start. You asked ChatGPT for a dentist in your town, you did not find yourself, and you took that as a verdict on your practice. AI passed no verdict. It never looked at your dentistry at all. It searched for things written about you, found very little, and named practices it could describe instead.

In other words, your problem is not marketing. It is record-keeping. Record-keeping is dull, cheap and slow. No campaign wins it. A dental practice wins it by writing down what it does, in the places that patients and machines both read, and then keeping those records true.

In practice, that means you can stop defending your star rating and stop refreshing ChatGPT to see whether its mood has changed. Ask your patients for reviews. Write a page for every town and every treatment you cover. Make your Google profile say what your website says. Then keep it true. That is the whole job.

If you are not sure how to do any of that, write to me.