Someone opens the ChatGPT app and types “dentist in Denver.” Instead of a list of links, they get about four dental practices shown as cards, each with a star rating and review count, and often a follow-up question about their area, needs or insurance.
We wanted to know where those dentists come from, what ChatGPT reads before it answers, and what agencies can do to make a client one of the dental practices it keeps recommending.
- ChatGPT picks dentists from Yelp and Google business listings. 96% of 3,901 recommendations came from a listing: 57% from Yelp, 39% from listings carrying Google Place IDs.
- It searches for the best-reviewed dentists first. Every one of 1,000 answers ran a web search, and 96% of them asked for the “best” or “top” dentists, reviews, or both.
- It reads directories and review sites, not practice websites. Healthgrades appeared in 73% of answers; Birdeye, Zocdoc and WebMD in about 58%. Practice websites were 15% of the pages it read.
- A high star rating beats a long review count. 87% of the dentists it recommended were rated 4.5 stars or higher, yet almost a third had fewer than 50 reviews.
- Most cities have a dentist it picks almost every time. Each city’s top dentist appeared in 92% of that city’s answers on average.
- Patients see cards, not sources. ChatGPT named about 4 dentists per answer, cited sources in none of 1,000 answers, and asked a follow-up question in 55%.
To learn how ChatGPT picks a dentist, we asked the ChatGPT app for one 1,000 times.
AI answers change from one run to the next, so a single answer is an anecdote. We asked the signed-in ChatGPT app the same unbranded question, “dentist in [city]”, 20 times in each of the 50 largest US cities. Every answer ran in a fresh temporary chat, so no memory, history or custom instructions shaped it. A browser inspector recorded what ChatGPT searched, every page it read and every business listing behind its answer.
Signed in only. These results come from a signed-in ChatGPT account. Signed-out ChatGPT shows no data to inspect, so we don’t yet know whether patients who aren’t logged in get the same listings-based answers. We’re testing that next.
Dental practices are anonymized throughout. The findings are about the playing field, not any one player.
ChatGPT builds its dentist list from Yelp and Google business listings.
Nearly every recommended dentist came with a business listing, Yelp most often
Each listing gives ChatGPT the dental practice’s name, address, phone number, star rating and review count. That is what fills the dentist cards and map in its answer. 87% of the dentists it recommended were rated 4.5 stars or higher, and only 1% were below 4.0. Review counts mattered less: the median was 147 reviews, and almost a third had fewer than 50.
Before recommending a dentist, ChatGPT searches for the best-reviewed.
The patient never asked for the “best” dentist, but ChatGPT added it anyway. It treats a plain “dentist in [city]” question as a search for the best-reviewed dentist.
ChatGPT reads dentist directories and review sites far more than dental practice websites.
The same eight sites appear in city after city
Of the 12,230 pages it read, 41% were dentist directories, 18% ranking and “best of” sites, 12% review platforms and 15% dental practice websites. Practice websites showed up in about a third of answers, usually one or two among a dozen pages. The ADA directory appeared in 11% of answers and Reddit in 2%; Yelp’s own pages never did. Yelp mattered only through its listing data.
ChatGPT recommends the same few dentists in each city.
Each row is one dental practice: the dentist ChatGPT recommended most in that city
| #1 dentist inone practice per row | Appeared inof the city’s 20 answers | Its ratingstars | Its reviewson that listing | Listing ChatGPT usedYelp or Google |
|---|---|---|---|---|
| Chicago | 20 of 20 | 4.7 ★ | 75 | Yelp |
| Seattle | 20 of 20 | 4.7 ★ | 53 | Yelp |
| Los Angeles | 20 of 20 | 4.9 ★ | 126 | Yelp |
| Dallas | 20 of 20 | 4.9 ★ | 40 | Yelp |
| Las Vegas | 20 of 20 | 4.9 ★ | 511 | |
| Houston | 19 of 20 | 5.0 ★ | 517 | |
| Tucson | 19 of 20 | 5.0 ★ | 680 | |
| San Francisco | 15 of 20 | 4.8 ★ | 141 | Yelp |
| Washington, DC | 13 of 20 | 4.9 ★ | 336 | |
| New York | 12 of 20 | 5.0 ★ | 22 | Yelp |
Across 20 answers, a city typically saw 6 to 19 different dentists (12 on average): only 6 in Chicago and Seattle, but 16 in New York and 18 in Washington, DC. The top one appeared in 92% of that city’s answers. In 46 of 50 cities, at least one dentist appeared in 80% or more of answers.
Most dental practices have both a Yelp and a Google listing; “Listing ChatGPT used” is the one ChatGPT pulled for its answer. The rating and review count are from that listing only. Every top dentist is rated 4.7 stars or higher, but review counts range from 22 to 680, so a high rating mattered more than a long review history.
ChatGPT shows about four dentist cards, cites no sources, and often asks a follow-up.
Each card is filled from the dental practice’s listing: name, address, star rating and review count. That listing is the first thing a patient sees, before any website. The pages ChatGPT read stay behind the scenes, so a practice can’t tell from the answer which directory or review site helped it.
ChatGPT recommends the highly rated dental practices it finds in Yelp and Google business listings. It searches for the best-reviewed dentists, checks them against directories and review sites, and shows each pick as a card built from its listing. The dental practice’s own website plays a minor role.
Seven ways to get a dental practice recommended by ChatGPT.
- 01
Claim and complete the Yelp listing
Accurate name, address, phone, hours and services, with photos and a steady flow of genuine reviews.
Why57% of the dentists ChatGPT recommended came from Yelp listings, more than any other source. - 02
Keep the Google Business Profile accurate
The same details as Yelp, kept in sync, with reviews answered.
Why39% of recommendations came from listings carrying Google Place IDs. - 03
Protect a 4.5-star rating
Ask every happy patient for a review and respond to the unhappy ones. Rating matters more than review volume.
Why87% of recommended dentists were rated 4.5 stars or higher and only 1% were below 4.0, yet almost a third had fewer than 50 reviews. - 04
Complete the directory profiles ChatGPT reads
Healthgrades, Birdeye, Zocdoc and WebMD: claim each profile and keep it complete and consistent with the listings.
WhyChatGPT read these sites in 58% to 73% of answers. - 05
Earn a place on the ranking and review sites
DentaScore, TopRatedDentist, Opencare, Yably and local “best of” lists.
WhyChatGPT read these in 40% to 57% of answers, in city after city. - 06
Keep the website accurate, but don’t lead with it
Make sure the homepage matches the listings. Put the bigger effort into listings, ratings and directories.
WhyPractice websites were 15% of the pages ChatGPT read, in about a third of answers. - 07
Measure it the way patients ask
Ask the signed-in ChatGPT app “dentist in [city]” in a fresh temporary chat, many times, and record which listing (Yelp or Google) each recommendation shows.
WhyAnswers vary: across 20 answers, a city saw 6 to 19 different dentists.
How we measured ChatGPT’s dental recommendations, and what the data can’t tell us.
1,000 answers from the signed-in ChatGPT app, collected October 14–28, 2026: “dentist in [city]” 20 times in each of the 50 largest US cities, each in a new temporary chat with no memory or custom instructions sent. Two answers that used a different prompt were excluded, and 14 extra answers beyond 20 in a city were set aside so every city counts equally.
565 answers ran on gpt-6-mini and 435 on gpt-6. Both named about 3.9 dentists per answer; gpt-6-mini asked a follow-up question more often (62% vs 45%). Where both models ran in a city, their top 3 dentists overlapped by about two.
A browser extension read the data ChatGPT sent to the browser: searches, pages read, business listings and the dentists named. Dental practices are anonymized.
The process was consistent across all 1,000 answers, so the conclusions about how ChatGPT works are solid. With 20 answers per city, an individual dentist’s rate carries up to about ±20 points of uncertainty. Signed-out ChatGPT is still being tested.