AI dentist recommendations are inconsistent: ask ChatGPT, Gemini and Claude the same question about the same city and they mostly name different practices. A study published in October 2026 found that of 560 practices recommended across the three assistants, only 30 appeared on all three lists.
For a practice owner, that changes what “being visible” means. There is no single ranking to climb. There is a spread of answers that shifts by assistant, by city and by the day you ask.
What did the study find about AI dentist recommendations?
Dental.me, a national dental directory, asked ChatGPT, Gemini and Claude the same question, “Who are the best dentists there?”, in each of the 50 most populous U.S. cities on October 16, 2026. Each query ran in a fresh session with no history, and a second round in the 10 largest cities tested repeatability. Its press release reports the numbers below.
| Measure | Result |
|---|---|
| Distinct practices named across all three assistants | 560 |
| Practices named by all three | 30 |
| Practices named by only one assistant | 76% |
| Cities where no practice appeared on all three lists | 28 of 50 |
| Recommendations that changed on a repeat query (10 largest cities) | 38% |
One caveat belongs next to these figures. Dental.me sells directory data and its listings were cited by ChatGPT and Claude in 11 cities, so it has a stake in the story. The study also measures who gets recommended, not who provides better care.
How rare is a practice that all three assistants agree on?
Rarer than the headline suggests. Dividing 30 by 560 gives about 5.4%, so roughly one named practice in 19 earned a mention from every assistant. If 76% were named by a single assistant, that is around 425 practices with one vote each, which leaves only about 100 practices that two assistants agreed on.
The useful reading is that the assistants behave less like a ranking and more like three independent samples from a much larger pool of plausible answers. A practice that appears in one sample today may not appear in the next. Agreement across assistants is the exception, so a practice cannot judge its visibility from one screenshot of one chatbot.
Why do AI assistants recommend weaker practices?
The study reports one example that stands out. In Baltimore, Gemini recommended a practice with a 3.1-star Google rating from 24 reviews, while Dental.me’s data lists 49 Baltimore practices rated 4.8 or higher with more than 100 reviews. Star ratings were not what decided that answer.
The assistants build answers from whatever sources they can read: directory pages, a practice’s own website, listings, news mentions and review text. A practice with thin or inconsistent information gives them little to work with, and a thinner profile can still win the mention if the assistant happens to retrieve it. Google’s own documentation says the usual search fundamentals still apply for its AI features in Search: helpful, crawlable content with no special markup required. My reading is that the same logic holds for the other assistants. Clear, consistent, easy-to-quote facts about a practice get picked up more often than a strong reputation that is hard to find.
How can a practice test its own AI visibility?
Because answers vary, a single check proves nothing. A small repeatable test works better than one lucky screenshot.
- Write five prompts a real patient would type, such as “best dentist in [city]”, “emergency dentist near [neighborhood]”, “who does Invisalign in [city]” and “dentist that takes [insurance] in [city]”.
- Run each prompt in ChatGPT, Gemini and Claude, using a fresh or temporary session so past chats do not shape the result.
- Repeat the same prompts on three different days, since the study saw 38% of recommendations change on a repeat query.
- Log each run in a sheet: assistant, prompt, date, whether your practice was named, which competitors were named, and which sources were cited.
- Count appearances out of total runs. Fifteen prompts across three assistants and three days gives 135 runs, enough to see a pattern.
My position is that appearance rate is the metric to track, not rank. Being named in 12 of 135 runs and then 31 of 135 after a round of fixes is a real result. Being “number one” in one chat is noise.
What should a practice fix when the results are patchy?
Start with the sources the assistants cited in your log, because those show where the answers came from. Then work through the profile data they depend on.
- Consistent facts everywhere. Name, address, phone, hours and services should match across your website, Google listing and major directories. Mismatches give an assistant two versions of the same practice to choose between.
- Service-level pages. A page per treatment, with the plain answers patients ask (who it suits, what it involves, which insurance applies), gives a retrievable passage to quote. Our guide to AI search for medical practices covers how these pages fit together.
- Review detail, not just review count. Reviews that mention specific treatments and staff are more useful to an assistant than a bare star average. The Google Business Profile guide explains why review recency matters now.
- Presence in the directories that get cited. If your log shows the same directory appearing in answers, make sure your listing there is claimed and complete.
None of this guarantees a mention, and the study’s own repeat results show why. It improves the odds across many runs, which is the only thing that can be improved.
What mistakes make AI visibility testing misleading?
Testing once and drawing a conclusion is the most common one. A second is asking with your practice name in the prompt, which tests whether the assistant knows you exist, not whether it recommends you to a stranger. A third is testing only the city name when patients search by neighborhood, treatment and insurance. Logged-in sessions can also carry earlier context, which is why the study used fresh or temporary chats.
FAQ
Do ChatGPT, Gemini and Claude give the same dentist recommendations?
No. In the Dental.me study, 76% of recommended practices were named by only one of the three assistants, and in 28 of 50 cities no practice appeared on all three lists.
Are AI dentist recommendations stable over time?
Not reliably. When the study repeated its query in the 10 largest cities, 38% of the recommendations changed, and Gemini’s list for Chicago was replaced entirely.
Does a high Google rating guarantee an AI recommendation?
No. The study found Gemini naming a Baltimore practice rated 3.1 stars from 24 reviews while dozens of 4.8-star practices with over 100 reviews were available.
How often should a practice check its AI visibility?
A monthly run of the same prompts is enough to see a trend. Compare appearance rates month to month instead of reacting to a single answer.
Is the Dental.me study independent?
It was run by a dental directory that benefits from being cited by AI assistants, so the findings are best read as a useful signal and checked against the published methodology.
AI assistants are sampling from a wide pool of practices, and the practices that show up most often are usually the ones whose basic facts are easiest to find and confirm.
