Asking AI for a dentist? It mostly repeats Google

ChatGPT, Gemini, Perplexity and Google's AI Mode will all answer the question "who is the best dentist in my town?" with a short list of names. This study looks at which practices make that list.

We wanted to know who gets named, and what the named practices have in common. So we drew 81 towns across England at random, nine in each region, from small market towns with a handful of practices to London boroughs with dozens. We asked four assistants two plain questions about each one, once each, and recorded every practice they named. Four towns were then left out because the Google Maps check could not be made reliably there, which leaves 77.

One caution comes first. ChatGPT, Gemini and Perplexity were asked through their developer versions (gpt-5.4-mini, gemini-3.8-flash and Perplexity's sonar), which are smaller models than the consumer apps may use. The apps also add the user's location, memory and map cards. Google AI Mode was read as the public sees it, set to the town, so it is the closest of the four to what a patient sees.

Three terms are used throughout. Named means the answer gave the practice's name. The Maps top three means the first three practices Google Maps listed for "dentist" and the town's name, searched from that town on the same day. Reviews means the number of Google reviews on the practice's Maps listing that day.

1. The Google Maps top three are named most, but not only them

A practice in the Maps top three was named in 59% of answers about its town. A practice ranked 4th to 10th was named in 25%, and one ranked 11th to 20th in 10%. The practice Maps put first was named 67% of the time.

Every assistant favours the Maps top three. None names only them.

Google AI Mode, Google's own assistant, did not follow Google Maps more closely than the others. It named a top-three practice in 57% of cases, against 69% for Gemini, 60% for Perplexity and 48% for ChatGPT. ChatGPT gave the shortest lists, 3 practices an answer on average, against 5 for Gemini and 4 for Perplexity and AI Mode, which is why its rates are lower across the board.

The other side matters as much. Of the times an assistant named a practice from the town's Maps results, 45% were outside the top three. Counted by practice rather than by mention, 59% of the Maps practices named at least once were outside the top three. The first practice from the town's Maps results that an answer named was a top-three practice 66% of the time, so about a third of first picks went to a practice Maps had lower down.

Town size changes the odds, so we checked it. In towns with fewer than 6 practices within 3 km of the centre, a top-three practice was named 75% of the time and the rest 20%. In towns with 15 or more, it was 36% against 18%. With fewer practices to choose from, every practice is more likely to be named, but the top three stay ahead in every size of town.

2. More reviews go with being named, inside the top three and out of it

Practices with 300 or more Google reviews were named in 43% of answers about their town. Those with 100 to 299 were named in 31%, 20 to 99 in 30%, and under 20 in 11%. The median named practice had 192 reviews. The median practice not named had 99.

Reviews and Maps position go together, so the obvious question is whether reviews only matter because they lift a practice up Maps. It does not look that way. Outside the top three, practices with 300 or more reviews were named 29% of the time, against 23% for 100 to 299, 18% for 20 to 99 and 6% for under 20.

Outside the Maps top three, the more reviews a practice has, the more often it is named.

To pull the strands apart we compared each practice only with the other practices in the same town's Maps results, answer by answer, so that town size, region and the assistant all drop out. On that comparison, every tenfold rise in reviews, say from 30 to 300, went with a rise of 18 points in the chance of being named (95% range 15 to 22), alongside 18 points for a top-three place (range 12 to 25). Using the exact Maps position instead of the top-three line gave 17 points for reviews. That comparison leaves out the 100 practices we could not class as independent or group. Keeping them in, without the group term, gives 16 points. The link to reviews was strongest for Google AI Mode (26 points) and weakest for ChatGPT (11), but present for all four. For ChatGPT, a top-three place counted for more than reviews did.

The star rating made no difference overall. Among practices with at least 20 reviews, those rated under 4.5 were named 31% of the time, 4.5 to 4.7 37% and 4.8 or more 35%. The assistants pulled in different directions: once reviews and position were allowed for, AI Mode leaned towards higher-rated practices, by about 10 points, while ChatGPT and Perplexity leaned away from them, ChatGPT by about 9. So, overall, the rating made no difference.

3. Independents are named as often as groups, with half the reviews

We used the CQC register to tell group practices from independents. A practice counts as part of a group if its CQC provider runs four or more dental locations, or its website is a known group's. Of the 792 Maps practices, 477 were independent, 215 were in a group and 100 could not be matched to the register with confidence, so are left out of this comparison.

Independents were named in 33% of answers about their town and group practices in 32%. That near-tie hides a difference. The median group practice had 242 Google reviews and the median independent 106. At like-for-like reviews and Maps position, independents were named at least as often as groups, ahead by between about 5 and 9 points depending on how branch names are counted.

That range comes from how groups are named. An answer that gives a group's brand name without saying which branch cannot be credited to any one branch, and in our main count we did not credit it. Counting such a mention as naming every branch of that group in the town instead narrows the gap from about 9 points to about 5. Either way the assistants show no lean towards groups.

4. The four assistants rarely agree

We put the same question about the same town to all four. Of the Maps practices any of them named, 34% were named by one assistant only, and 18% by all four. In 10% of 144 questions all four put the same Maps practice first.

A third of the practices named by one assistant are missing from the other three.

Each assistant's two answers about the same town also differed. Asked "what's the best dentist" and "I need a new dentist, who do you recommend", the overlap between its two lists, counted as the practices on both out of the practices on either, was 71% for Gemini, 64% for Perplexity, 55% for AI Mode and 38% for ChatGPT. Each question was asked once, so we cannot say how much of that comes from the wording and how much from an assistant giving a different answer each time it is asked.

37% of the Maps practices in our towns were not named by any assistant, in any answer.

5. Almost every practice named is a real one nearby

Of the 2,431 practice names the assistants gave, 81% matched a practice in the town's Google Maps results. 11% matched another CQC-registered practice within 10 km, in the town or nearby but not in its Maps top 20 under that name. 3% were a group's brand name that fitted more than one branch nearby. That makes 95%. A further 3% matched the name of a registered practice elsewhere in England.

That leaves 71 names we could not match automatically. Only names an answer set apart, in bold, a heading or a Google place card, were collected for this check. Names given in plain text that matched no practice were not. 32 of them share a distinctive word with a registered practice within 25 km, which usually means a different form of the same name. We checked the rest by hand against the CQC register. 16 turned out to be a real practice under a near-identical or older name, and 2 were too generic to trace, such as "The Dental Practice". 21 mentions, about 1% of all names given, we could not trace at all. That does not show those practices do not exist. The register lists practices by their legal names, which can differ from the name on the door, so we have not called any of them invented.

What to do about it

Ask the assistants yourself, more than once. Ask ChatGPT, Gemini and Perplexity, and search in Google's AI Mode, for the best dentist in your town, in two or three different wordings, and on more than one day. Only 18% of the practices named were named by all four, so a single answer tells you little.

Work on your Google reviews. Reviews and Maps position were the two signals associated with being named, for every assistant and every size of town. Reviews are the one a practice can most directly work on. We cannot show that more reviews cause more mentions, but the link held inside and outside the Maps top three. The count mattered and the rating did not. A steady habit of asking every patient for a review is the most practical step we can point to.

Keep your Google listing linked to your website. In the comparison within each town, a listing with a website went with a rise of 13 points in the chance of being named (95% range 7 to 19). That rests on a small group: only 104 listings had no website, and 70 of those had under 20 reviews. What the website does is give the assistants something to read. Perplexity cited a practice's own website in 86% of its answers and ChatGPT in 33%. Make sure yours says plainly what you do, where you are and whether you are taking new patients.

Check the pages the assistants read about you. The NHS website was the source ChatGPT and Perplexity cited most often, followed by dental directories and the CQC. If you have an NHS website profile, check that its name, address, services and new-patient status are right, and do the same for the larger directories once a year.

Independents have no reason to hold back. At like-for-like reviews and Maps position, independents were named at least as often as group practices.

How we did this

Towns. We took Ordnance Survey towns and cities in England, and London boroughs, that have a unique name in our town list, a Google local result for "dentist" and the town, and at least three CQC-registered dental locations within 3 km of the centre. In each of England's nine regions we split the towns into thirds by that count and drew three from each third at random, with a fixed seed, giving 81 towns. Four were then left out: for two the Google Maps search landed on a different place of the same name, and two had fewer than three practices in their Maps results. The median town in the study has 9 practices in its Maps results. London boroughs have no centre point in our town list, so we used the average position of each borough's CQC dental locations.

Questions. Each assistant was asked "What's the best dentist in <town>, <county>?" and "I need a new dentist in <town>, <county>. Who do you recommend?" The county is there so the assistant answers about the right place: there is a Newport in Shropshire, one in Wales and one on the Isle of Wight. ChatGPT (gpt-5.4-mini), Gemini (gemini-3.8-flash) and Perplexity (sonar) were asked through the developer versions offered by DataForSEO's AI Optimization API, with web search on. Google AI Mode was read through DataForSEO's Google AI Mode results, set to the town. All 616 answers were collected on 23 October 2026. A patient's own app also knows their location and history and may run a larger model, so their answer can differ. This is the answer each assistant's developer version gives, and Google AI Mode's answer for the town.

Google Maps. On the same day we took the top 20 Google Maps results for "dentist" and the town, searched from the town. Listings that were not dental, NHS community services, listings more than 20 km from the town centre, and listings for an individual dentist at a practice's address were set aside or folded into the practice, which left 792 practices. We also checked the top three from an earlier Google search on 13 September 2026: practices in that top three were named 62% of the time, against 19% for the rest, the same pattern.

Matching names. A practice counts as named if its name appears in the answer as whole words, in full, cut at its first comma or dash, under a former name it gives in brackets, or without the town at the end, or if a practice name the answer sets apart (in bold, in a heading, or on a Google place card) has the same distinctive words as the listing, checked both ways round. Apostrophes and spacing are ignored. Generic words such as "dental" and "practice" and the town's own name are not enough on their own. A name that fitted more than one practice in the town, or a group's brand without its branch, was credited to none. Names that matched nothing in the town were checked against every CQC dental location within 10 km, then against the whole register. Of 40 matches drawn at random and read by hand, all 40 were right, 4 of them on a near-identical name we judged to be the same practice. The test errs towards missing a practice rather than crediting the wrong one.

Groups and independents. Each Maps practice was matched to the CQC care directory of 14 October 2026 by name and location, or by postcode. Where a site carries two current registrations while it changes hands, the larger provider was used. Four or more dental locations under one provider, or a website belonging to a known group, counts as a group.

The comparison within a town. The 18 point figure comes from a linear model of whether each practice was named, with a separate baseline for every answer, so each practice is only ever compared with the others in its own town's results for the same answer. It includes the number of reviews (on a tenfold scale), a top-three place, having a website, being in a group and a rating of 4.8 or more, and leaves out the 100 practices whose group status is unknown. The ranges are 95% intervals from resampling the towns 1,000 times.

Controls we ran. Each question on its own, London apart from the rest, towns apart from cities and boroughs, each third of town size, and the exact Maps position in place of the top-three line. Reviews and Maps position were associated with being named in all of them, and group practices came out behind independents in all of them. Splitting each town's practices into its better-reviewed and less-reviewed halves, the better-reviewed half was named in 51% of answers and the other half in 20%.

What this is not. It is one day's answers, with one ask per question, and the assistants' answers can change. It shows which practices are associated with being named, not what causes it. It covers England only, because the CQC register does. No practice or town is named in this study or in its data file.

Questions we get asked about this data

Do AI assistants recommend the same dentists as Google Maps?

Mostly, but not only. A practice in the Google Maps top three for its town was named in 59% of answers, against 20% for practices ranked 4th to 20th. Of the times an assistant named a practice from the town's Maps results, 45% were outside the top three.

Do Google reviews affect whether ChatGPT recommends my practice?

The number of Google reviews was associated with being named by every assistant we tested, ChatGPT included, even after allowing for Maps position and town. That is an association, not proof of cause, and for ChatGPT a top-three place on Maps counted for more. The star rating made no difference overall once a practice had 20 or more reviews.

Do AI assistants invent dental practices?

We found little sign of it. 95% of the names given matched a practice on Google Maps or the CQC register near the town. About 1% could not be traced at all, and we have not called any of them invented, because a name missing from the register is not proof that the practice does not exist.

For editors

The pooled figures are free to reuse with attribution to Practice Radar, under Creative Commons BY 4.0. Sources: answers from the developer versions of ChatGPT, Gemini and Perplexity and from Google AI Mode, and Google Maps results, collected through DataForSEO on 23 October 2026, and the CQC care directory of 14 October 2026. The data file, ai-assistants-recommend-dentists.csv, has one row per assistant with the rates in this article. It names no practice and no town.

Jack Henty, who compiled this analysis, co-owns Bollington Dental Practice in Cheshire and founded Practice Radar, which measures local dental markets for practices. He is available for interview.

Contact: hello@practiceradar.co.uk