We asked four AI tools for the best dentist in 20 US cities. Here is who they picked and why

A patient types “Who is the best dentist in Raleigh?” into an AI tool. Which practices come back, and why those ones? We wanted real numbers rather than opinions, so in November 2026 we asked that question for 20 mid-size US cities, in four AI engines, twice each.

The short version: the four engines rarely agreed with each other, most of them often disagreed with themselves, and they read very different sources. For a practice owner, that changes where effort should go. Below are the numbers, the method and its limits, and what we think the patterns mean.

Key findings

All numbers are from Cleareo’s tests, November 2026:

• 489 different practices were named across 160 answers in 20 cities.
• Only 5 practices were named by all four engines. 425 were named by just one engine, 51 by two and 8 by three.
• Every city had at least one practice named by two or more engines. Across the 20 cities that was 64 practices.
• Claude named 10.1 practices per answer on average, Gemini 5.2, Perplexity 2.4 and ChatGPT 1.7.
• Perplexity was the steadiest: same first pick in 19 of 20 cities when re-asked, with 67% of practices repeated. Gemini kept its first pick in 12 cities, Claude in 8 and ChatGPT in 7.
• Claude quoted star ratings or review counts in 38 of 40 answers, Perplexity in 35, ChatGPT in 24 and Gemini in 5.
• The engines read different things. Directories and review sites were 55% of Claude’s source links and 49% of Perplexity’s. Practice websites were 46% of Gemini’s. Reddit threads were the most common type in ChatGPT’s search results (31%), but its footnotes pointed mostly to DentistsRanked and Healthgrades.
• 159 of 160 answers mentioned insurance.

How we ran the study

We asked one question, “Who is the best dentist in [city]?”, for 20 US cities: Austin TX, Boise ID, Charlotte NC, Columbus OH, Denver CO, Indianapolis IN, Kansas City MO, Madison WI, Minneapolis MN, Nashville TN, Omaha NE, Phoenix AZ, Pittsburgh PA, Plano TX, Portland OR, Raleigh NC, Richmond VA, Sacramento CA, San Diego CA and Tampa FL.

Each question went to four engines through their APIs, with web search switched on and a US location:

• ChatGPT: OpenAI’s Responses API with web search, on gpt-6-luna, the cheaper API model
• Claude: claude-sonnet-5-5 with Anthropic’s web search tool
• Google Gemini: gemini-3.5-flash-lite with Google Search grounding
• Perplexity: Agent API, “fast” preset

Gemini and Perplexity ran on November 8, 2026. Claude and ChatGPT ran on November 9. Each engine got each question twice, a few minutes apart. That is 40 answers per engine and 160 in total, and every answer used web search. The whole run cost about $4.34 in API fees: $3.21 for Claude, $1.07 for ChatGPT and about $0.06 for Gemini and Perplexity together.

The ChatGPT app may use a larger model than gpt-6-luna. As a check, we ran 4 of the cities against gpt-6.1-terra, and it gave the same first pick in 3 of the 4.

A second model pass pulled out every practice named in each answer, and we checked samples of that extraction by hand. We counted a practice once per city, matching on its name. For sources, we counted each URL once per answer, including both the links an engine footnoted in its text and the search results it returned alongside the answer. That matters for comparisons:

• ChatGPT returned 2,066 links, of which 77 were footnotes in the text.
• Perplexity returned 400, of which 100 were footnotes.
• Gemini returned 256, all of them cited links.
• Claude returned 368 search results. Our capture did not record which of them were inline citations, so all 368 count as read.

We sorted every source domain into one of six types (health directories and reviews, ranking and best-of sites, practice websites, Reddit and forums, local press and magazines, other) with one shared set of domain rules for all four engines, checked by hand. Find-a-dentist listings from insurers and state programs count as directories. State dental boards, dental associations and dental schools count as other. Practice websites means a single practice’s or dental group’s own site.

Limits you should know about

This is a small snapshot, not a census.

• API, not the consumer apps. Our earlier tests found that API answers differ from what people see in the ChatGPT, Claude, Gemini or Perplexity apps. They mostly kept the app’s top pick but changed a good share of the shortlist and the sources. ChatGPT also ran on a cheaper model than the app may use. Treat these numbers as a close relative of what patients see, not a copy.
• Two days, two runs per question, US only, mid-size cities only. Gemini and Perplexity answered a day before Claude and ChatGPT.
• One phrasing of the question. “Best family dentist near me” or “dentist that takes Delta Dental” would pull different results.
• Name matching is imperfect. Two branches of one group, or a practice listed under a slightly different name, may be counted separately.
• In Columbus, Claude named individual dentists from Healthgrades rather than practices, so our count shows zero Claude practices there.
• Google AI Overviews and AI Mode were not tested.

How many dentists did each AI name?

Claude named the most practices by far, and ChatGPT the fewest. Perplexity gave short, stable answers built on a few lists. Claude and Gemini gave longer lists that changed between runs. ChatGPT usually named one to three practices, and often different ones each time.

Measure	ChatGPT	Claude	Gemini	Perplexity
Answers	40	40	40	40
Practices named per answer (average)	1.7	10.1	5.2	2.4
Practices named across both runs, summed over 20 cities	56	301	155	59
Repeat overlap of practices, same question minutes apart	29%	38%	37%	67%
Same first pick on both runs	7 of 20 cities	8 of 20 cities	12 of 20 cities	19 of 20 cities
Answers quoting star ratings or review counts	24 of 40	38 of 40	5 of 40	35 of 40
Source links (URL counted once per answer)	2,066	368	256	400
Different websites among those links	422	86	106	63
Share of source links that were practice websites	9%	14%	46%	14%

“Repeat overlap” is the share of practices named in either run that were named in both, averaged across the 20 cities.

ChatGPT read far more than it named. It pulled in about 52 links per answer, more than five times any other engine, and then named fewer practices than any of them.

Did the four engines agree on the best dentist?

Rarely. Of the 489 practices named, 425 (87%) came from one engine only. Just 5 were named by all four, and 8 more by three. If a patient checks one AI tool against another, they will usually get a different shortlist.

The five practices all four engines named were Artisan Dental in Madison, Biltmore Commons Dental Care in Phoenix, D2O Dental in Sacramento, Laurelhurst Dentistry in Portland and Parkway Dental Center in Minneapolis. We mention them as examples of agreement, not as a ranking.

Some pairs of engines agreed more than others. Counting practices named by both engines in the same city:

Pair of engines	Practices both named
Claude and Gemini	36
ChatGPT and Perplexity	17
ChatGPT and Gemini	16
ChatGPT and Claude	15
Gemini and Perplexity	11
Claude and Perplexity	10

Claude and Gemini gave the longest lists, which helps explain why they overlapped most.

Results by city

The table shows how many practices each engine named across its two runs in each city, how many different practices the four named in total, and how many were named by two or more engines.

City	ChatGPT	Claude	Gemini	Perplexity	All practices named	Named by 2+ engines
Austin, TX	4	15	9	3	27	4
Boise, ID	2	15	11	4	24	7
Charlotte, NC	3	13	9	3	24	4
Columbus, OH	4	0	8	2	12	2
Denver, CO	1	13	7	2	22	1
Indianapolis, IN	3	13	7	3	22	4
Kansas City, MO	3	19	9	2	30	3
Madison, WI	3	15	7	3	22	4
Minneapolis, MN	3	17	8	2	23	4
Nashville, TN	3	19	5	3	24	5
Omaha, NE	3	18	7	2	27	3
Phoenix, AZ	1	15	7	2	19	3
Pittsburgh, PA	2	15	8	4	25	3
Plano, TX	3	14	10	5	31	1
Portland, OR	4	17	7	3	25	4
Raleigh, NC	3	18	7	4	31	1
Richmond, VA	2	16	6	2	23	2
Sacramento, CA	5	18	5	3	22	6
San Diego, CA	1	13	8	3	24	1
Tampa, FL	3	18	10	4	32	2
Total	56	301	155	59	489	64

Boise had the most agreement, with 7 practices named by two or more engines. Denver, Plano, Raleigh and San Diego had just one each.

This lines up with other research. A Dr. Bicuspid study found ChatGPT, Gemini and Claude named 560 different practices across 50 cities, with only 30 on all three lists. BrightLocal’s study of 200,000 AI answers found re-asking the same prompt brings back the same business only about half the time.

An example: Austin, TX

Perplexity named Blunn Creek Family Dentistry first in both runs, citing a ranking site’s figure of 4.9 stars from 182 reviews. ChatGPT named Blunn Creek alone in its first run, then three different practices in its second.

Gemini and Claude both named Forest Family Dentistry and Smile 360. Claude and ChatGPT both named ATX Family Dental. That is four practices shared by two engines, out of 27 named in Austin, and none named by three. Same city, same question, four different shortlists.

What sources did the AI read?

Each engine leaned on a different kind of page. Claude and Perplexity read mostly directories and review sites. Gemini read mostly practice websites. ChatGPT’s searches returned a wide spread, with Reddit threads the most common type, while its footnotes went mostly to DentistsRanked and Healthgrades.

Source type	ChatGPT links	Share	Claude links	Share	Gemini links	Share	Perplexity links	Share
Health directories and reviews	311	15%	204	55%	62	24%	195	49%
Practice websites	193	9%	53	14%	117	46%	54	14%
Ranking and best-of sites	219	11%	58	16%	29	11%	113	28%
Reddit and forums	637	31%	0	0%	28	11%	0	0%
Local press and magazines	101	5%	7	2%	9	4%	28	7%
Other	605	29%	46	13%	11	4%	10	3%
Total	2,066	100%	368	100%	256	100%	400	100%

Each URL is counted once per answer. “Other” covers everything else, such as Wikipedia (69 ChatGPT links), Birdeye review pages, Demandforce listings, state dental board pages, dental association pages and dental schools.

The most-cited sites

These 15 sites had the most links across all 160 answers. In brackets: the number of answers (out of 40 per engine) that included at least one link to the site.

Site	Type	ChatGPT	Claude	Gemini	Perplexity	Total links
reddit.com	Forum	637 (40)	0	28 (23)	0	665
dentistsranked.com	Ranking site	94 (40)	2 (2)	5 (5)	42 (40)	143
doctor.webmd.com	Directory	48 (36)	28 (22)	2 (2)	46 (38)	124
yelp.com	Reviews	0	119 (40)	0	0	119
healthgrades.com	Directory	79 (40)	12 (12)	0	22 (14)	113
en.wikipedia.org	Other	69 (29)	8 (4)	0	0	77
zocdoc.com	Directory and booking	28 (19)	3 (3)	26 (25)	18 (16)	75
opencare.com	Directory	20 (20)	31 (31)	16 (16)	8 (8)	75
toprateddentist.com	Ranking site	20 (15)	11 (11)	17 (17)	18 (14)	66
dentascore.com	Ranking site	38 (30)	3 (2)	0	1 (1)	42
dental.me	Directory	3 (3)	3 (3)	0	34 (34)	40
findadentist.ada.org	ADA dentist finder	38 (25)	0	0	0	38
health.usnews.com	Directory	0	0	0	31 (15)	31
reviews.birdeye.com	Reviews (counted as other)	26 (18)	2 (2)	0	2 (2)	30
deltadental.com	Insurer’s dentist finder	6 (6)	2 (2)	14 (14)	4 (4)	26

Two details matter here. First, Claude was the only engine that read Yelp, and it did so in every answer, mentioning Yelp in all 40. Second, most of ChatGPT’s and Perplexity’s links were search results, not footnotes. Reddit threads came back in the search results for all 40 of ChatGPT’s answers, but none of its 77 footnotes pointed to Reddit. Its footnotes went mostly to DentistsRanked (in 23 answers) and Healthgrades (6). Perplexity footnoted DentistsRanked in 38 of 40 answers and Dental.me in 19. So the ranking pages were what got quoted, while Reddit, WebMD and Healthgrades were read in the background.

This matches the wider pattern. Yext’s research found listings make up 52.6% of AI citations in healthcare, led by WebMD, Vitals, Zocdoc and Healthgrades.

What did the answers say about choosing a dentist?

Insurance came up almost every time: 159 of 160 answers mentioned it. Most Claude, Gemini and Perplexity answers said there is no single best dentist and that it depends on the patient. ChatGPT’s answers were short (about 105 words on average) and mostly went straight to one or a few names.

Claude and Perplexity backed their picks with numbers: star rating, review count, sometimes emergency availability, nearly always taken from a directory, review or ranking page. Gemini quoted ratings least, and was the only engine that pointed patients to Reddit in its answer text, in 16 of 40 answers.

What the patterns mean for a practice owner

The practices that showed up were the ones that existed, with consistent details, on the pages each engine likes to read. Because the engines read different pages, one strong channel is not enough. The good news is that most of those pages are ones you can fix or influence yourself.

Your own website is read by every engine

Practice websites were 46% of Gemini’s source links, 14% of Claude’s and Perplexity’s, and 9% of ChatGPT’s. This is the one source you fully control. Pages that say plainly who you are, where you are, which dentists work there, which services you offer and which insurance you take give the engine something it can repeat. Being read is not the same as being picked, but a vague homepage gives the engine nothing to quote.

Your directory profiles are being read

Directories and review sites such as WebMD, Healthgrades, Yelp, Zocdoc, Opencare, Dental.me and US News made up 55% of Claude’s source links and 49% of Perplexity’s, and ChatGPT pulled Healthgrades into every answer. Many practices have auto-generated profiles on these sites that they never claimed. Claim the ones you can, check the name, address, phone, hours, insurance accepted and services, and make sure each dentist is listed under the practice. If a directory shows you as accepting no insurance, or lists a dentist who left two years ago, that is what an AI may repeat.

If Claude matters to you, check your Yelp listing too. It was the only engine that read Yelp, and it read it every time.

Getting onto ranking lists shapes the first pick

DentistsRanked appeared in all 40 Perplexity answers and all 40 ChatGPT answers, and was the site both engines footnoted most. TopRatedDentist, DentaScore, LocalPicks.ai, BestInHood and ThreeBestRated also showed up. These sites publish city-by-city lists of dentists. We have not verified how each one builds its lists or whether any of them charge for placement, so read each site’s own explanation before you pay anyone to be included. What we can say is that the ratings and review counts these pages show are often the numbers the engines quote back.

Reviews get quoted with counts

Claude quoted star ratings or review counts in 38 of 40 answers, Perplexity in 35 and ChatGPT in 24. “4.9 stars from 182 reviews” is the kind of line that makes it into an answer. A steady flow of genuine Google reviews matters because those totals get copied onto directory and ranking pages, which the AI then reads. Ask every patient, never offer incentives and never filter who you ask; Google’s review policy bans both.

ChatGPT’s searches turned up Reddit threads most often

ChatGPT’s searches returned 637 Reddit links, in all 40 of its answers, more than any other source type. None of its footnotes pointed to Reddit, though; those went mostly to DentistsRanked and Healthgrades. Gemini linked to Reddit in 23 answers. You cannot and should not plant posts there. But it is worth searching your city’s subreddit for “dentist” to see what locals say about you, because those threads are what turns up when an AI tool searches your city.

City magazine lists and insurer directories still count

Local press and magazine pages were a small share of links (101 for ChatGPT, 28 for Perplexity), but they are local, dated and trusted. If your city magazine runs a peer-nominated or reader-voted dentist list, find out how it works and when nominations open. Insurer directories showed up too: Delta Dental’s find-a-dentist pages were read 26 times, and ChatGPT read the ADA’s dentist finder 38 times. Keep each network listing current, and say on your own site which plans you take.

What to do this week

Start with the sources that showed up most, and work outward.

1. Ask two AI tools “Who is the best dentist in [your city]?” twice each, and note which sites they list as sources. Our guide to getting your practice recommended by ChatGPT walks through what to check.
2. Add or fix a page on your site for each dentist and each main service, with the plans you accept.
3. Claim and correct your WebMD, Healthgrades, Zocdoc, Opencare and Yelp profiles. Check insurance and dentist names first.
4. Look up your practice on DentistsRanked, TopRatedDentist and your city magazine’s list. Note what they say about you, without paying for anything yet.
5. Read what decides which practices AI recommends for the full list of levers, ranked by evidence.

If you would rather see where you stand first, our free AI check runs five patient questions for your town in ChatGPT, Google’s AI, Claude and Perplexity, twice each, and emails you the results in two business days. We’re new, so judge us on that.

How to cite this study

Cleareo, “We asked four AI tools for the best dentist in 20 US cities. Here is who they picked and why”, November 2026 (updated November 9, 2026), https://cleareo.io/blog/ai-best-dentist-study-20-cities/. Data: 160 answers from the ChatGPT (gpt-6-luna), Claude, Gemini and Perplexity APIs with web search, 20 US cities, two runs each.