AI Is Now Choosing People's Dentists, and It Can't Agree With Itself

• Three assistants, three different answers. 76% of the 560 practices were recommended by only one of the three assistants. Just 30 practices in the entire country were named by ChatGPT, Gemini and Claude alike.
• No consensus in 28 of 50 cities. In 28 of the largest cities, not one practice appeared on all three lists. In Dallas, Tucson and Virginia Beach, the three assistants did not share a single recommendation.
• Ask again, get a different answer. Asked the identical question a second time in the 10 largest cities, 38% of the recommended practices changed (ChatGPT 36%, Gemini 44%, Claude 33%). In Chicago, Gemini's second answer replaced all five of its first picks.
• "Best" is not always highly rated. Asked for the best dentists in Baltimore, Gemini answered with "5 highly-rated dental practices", one of which carries a 3.1-star Google rating from 24 reviews. Dental.me lists 49 Baltimore practices with ratings of 4.8 stars or higher and more than 100 reviews.
• When AI searched the web, it turned to Dental.me in 11 of 50 cities. ChatGPT and Claude cited Dental.me as a source for their recommendations in Atlanta, Boston, Charlotte, Dallas, Denver, Jacksonville, Kansas City, Miami, Sacramento, San Diego and Washington, D.C. See where.

This study measures what AI assistants recommend. It does not evaluate the clinical quality of any dental practice, and inclusion or exclusion from an assistant's list says nothing about a practice's care.

Assistants that search the web show the sites they drew on. In these 11 cities, ChatGPT or Claude listed a Dental.me page as a source for its dentist recommendations. Claude, with web search turned on, was asked the study question 8 times in every city; the Claude column shows how many of those answers drew on Dental.me.

For each city: how many different practices the three assistants named in total (each named up to five), how many were on all three lists or on two, and the practices that more than one assistant agreed on. Practices named by only one assistant are counted but not listed. The last column shows how many practices in that city Dental.me lists with 4.8+ stars and 100+ reviews.

In the 10 largest cities, each assistant was asked the identical question a second time on the same day. The table shows how many of the second answer's five practices also appeared in the first answer.

The question. Each assistant was asked, word for word: “I'm looking for a dentist in [City, State]. Who are the best dentists there? Please name 5 specific dental practices with their phone numbers.”, with each of the 50 most populous U.S. cities filled in.

The assistants. ChatGPT (chatgpt.com), Google Gemini (gemini.google.com) and Anthropic's Claude (claude.ai), using their standard consumer apps on October 16, 2026. Each question was asked in a fresh session with no prior history: ChatGPT temporary chat, Claude incognito chat and a new Gemini chat.

Matching. Practices were matched across assistants by phone number where both answers gave the same number, and otherwise by distinctive words in the practice name. Generic words such as “dental” or “downtown” were never enough on their own. A different phone number was not treated as proof of a different practice, because assistants sometimes give outdated numbers.

Sources cited. The consumer Claude app answers from map listings and does not show web sources, so to see which websites Claude draws on, researchers also asked the same question through Anthropic's API with web search turned on, 8 times in each of the 50 cities (400 answers), and recorded every source. ChatGPT's cited sources were read from its answers. Only links to dental.me itself were counted. Claude also drew on other sites, including review sites and other directories, in many answers.

Ratings. Star ratings and review counts cited here were checked against each practice's public Google listing on October 16, 2026.

Limits. AI answers change over time and can vary by location, account and device, so a reader repeating this test may see different results. That variability is the point of the study.