The engines recommend the Austin dentists they read

What did we find?

The engines recommend the dentists they read. Across two runs of 36 checks each, ten practices had their own website cited by ChatGPT, Gemini or Perplexity at least once, and all ten were named in answers. Ten practices never had their own website cited, and seven of the ten were never named in any of their 144 checks. The probability of a split that clean arising by chance is about 0.0015 (one-sided Fisher exact test, n = 20). It held in each run separately: 5 of 5 and 11 of 13 in run 1, 8 of 8 and 8 of 12 in run 2.

Seven of the twenty practices were never named in either run, across all twelve questions and all three engines. These are not obscure businesses: every one is among the twenty most reviewed dentists in Austin on Google Places. The most visible practices, Tech Ridge Dental, Walden Dental and Toothbar, were named in 14, 13 and 12 of their 72 checks. No practice was named in more than one check in five.

Figure 1Answers naming each practice, by whether its own website was cited

This repeats the central finding of our September study of 17 Central Texas trade businesses, in a different industry with a different citation landscape. There, own-site citation correlated with recommendation at rho +0.92. Dentistry is far more directory-dominated than roofing, and the pattern held anyway.

It is also consistent with Herringbone's September 2026 study of 600 businesses, which found 27% of Google AI Overviews answers contained details the business never published, and that accuracy jumped to 84–89% when the business states specifics plainly on its own site. Their question was whether AI gets your facts right; ours is whether it says your name at all. Both point at the same lever: a website that says plainly what you do, where, for whom, at what price.

Which dentists did AI name?

Practices were drawn from Google Places by searching dental practices in Austin and ranking by review count, not from lists the engines already favour. Engine errors, mostly Perplexity rate limits, are excluded from “measured” rather than counted as absences.

What do the engines actually read?

By citation volume, directories and portals dominate. Two dental directories most people have never heard of, opencare.com and toprateddentist.com, were cited 420 and 362 times across the two runs, well ahead of anything else. Reddit threads were cited 231 times, more than WebMD.

Figure 2Most cited domains across every measured answer

Volume and effect are different things. Directories are most of what the engines read; a practice's own website is what tracked being recommended. Both are true, and a report that states only one of them misleads.

Do Google reviews help you show up in AI answers?

Review count cannot be what separates these practices, because count is how they were selected: all twenty are the most reviewed in the city, 693 to 1,648 each, and the seven never named in either run span that whole range. Swish Dental holds four of the twenty spots, roughly 4,270 reviews across its Domain, Downtown, South and Mueller locations. The flagship at The Domain was never named in 53 measured answers, Mueller never in 55; Downtown and South were named 3 and 2 times in 72 checks each. Whatever gets a practice into an AI answer, it is not the number next to the stars.

Review quality may be a different matter, but this study cannot show it. Star rating correlated with how often a practice was named at Spearman +0.60 in run 1 and only +0.29 across both runs (n = 20), which is too weak to build on; the top three practices do all hold 4.9 or 5.0, and the only sub-4.0 practice, Brident at 3.7, was named twice in 72 checks. This does not mean reviews do not matter.The engines cited review platforms constantly, Healthgrades 76 times, Reddit 63, Zocdoc 61. What they take from those pages is text, not a tally. A review that says “great dentist, friendly staff” gives an engine nothing to reuse. One that says which dentist did which procedure at which location, took which insurance, and what problem it solved is a sentence an engine can quote.

The practical reading, offered as a hypothesis this study did not test directly: ask the customers you already have to say what you did, where, and what it fixed. That is the same lever as the website finding, applied to the pages you do not control.

Figure 3Google reviews against answers naming the practice, 20 practices, two runs

Who gets named instead?

When the twenty most reviewed practices are absent, someone else fills the answer. The names the engines reached for most often across both runs: Forest Family Dentistry, Lone Star Circle of Care, McLane Family Dental, Enamel Dentistry, Manos de Cristo Dental Center, CommUnityCare Health Centers, South Austin Dental Associates and Austin Artistic Dental, each named in 36 to 38 separate answers, more than twice as often as any practice in the sample.

Two things stand out in that list. First, McLane Family Dental and Austin Artistic Dental are the two practices whose websites appear in the top cited sources: the engines read them, then recommend them. Second, the affordability questions pull in community clinics: Manos de Cristo, Lone Star Circle of Care and CommUnityCare are non-profits with plain pricing and service pages, and on “affordable dentist in Austin” they, not the most reviewed private practices, are the answer.

How did we measure this?

We drew the twenty most reviewed dental practices in Austin from Google Places, ranked by review count at the time of sampling. For each, we asked the same twelve questions on ChatGPT, Gemini and Perplexity, 36 checks per practice per run, twice, on 19 and 21 September 2026, and classified every answer as recommended, mentioned but not recommended, used as a source but not named, or absent. We recorded every URL every engine cited. Business name, category and city were supplied explicitly for every check, never derived from page metadata.

This is the complete question set; only category and city are substituted:

• best dentist in Austin, TX
• who is the top rated dentist in Austin, TX
• recommend a good dentist near Austin, TX
• most reliable dentist in Austin, TX
• affordable dentist in Austin, TX
• who should I hire for dentist work in Austin, TX
• how do I choose a good dentist in Austin, TX
• what should I look for when hiring a dentist
• questions to ask a dentist before hiring
• which dentist companies serve Austin, TX
• compare dentist options in Austin, TX
• top 5 dentist businesses in Austin, TX

These are templates, not observed searches. No AI platform publishes what people type into it; there is no Search Console for ChatGPT. Anyone claiming volume-validated AI prompts is using Google data as a proxy for a different medium, or guessing. We are guessing too, and saying so. The templates exist for comparability: twenty practices asked identical questions produce scores that can be compared, and bespoke questions per practice would not.

An engine error, almost always a Perplexity rate limit, removes that check from the denominator rather than counting as an absence. A run where fewer than 60% of a practice's checks produced a usable answer is withheld from every conclusion and published as unmeasured; that withheld Enamel Dentistry Parmer Park and Austin Dental Company from run 1. Run 1 fired the twenty practices concurrently and Perplexity answered only 81 of 240 checks; run 2 fired them one at a time and Perplexity answered 235 of 240, so run 2 is the better-covered run and every practice is measured in it. Review counts and ratings are from Google Places at time of sampling.

What are the limitations of this study?

Twenty practices, one city, one profession, two days. Strong enough to show that own-site citation and being named travel together in this market, not enough to quantify what any single factor contributes, and not a claim about other cities or trades.

Correlation is not causation. A third factor, most plausibly a clear, crawlable website, may drive both own-site citation and being recommended. We report the split and the p-value, not a mechanism.

Engines vary between runs, and we measured how much. Of the eighteen practices measured in both runs, twelve kept their named-or-not status and six changed it: five went from never named in run 1 to named one to three times in run 2, and one went the other way. Most of that movement is coverage, not caprice: Perplexity answered three times as many checks in run 2 and it is the engine most willing to name a practice. Per-practice instrument scores moved by a median of 1 point and at most 8. Run 1 alone said 11 of 18 practices were never named; run 2 alone said 8 of 20; together, 7 of 20 were never named in 144 checks. The own-site split held in both runs on its own. We publish both runs in full rather than the tidier one. Our standard is a week between runs, and these two are two days apart, which we flag rather than hide; a third run a week out is planned and will be added here.

The sample is the most reviewed twenty, not all of Austin. Google Places pagination and its ranking of “dentist” results decide who makes the list, so a heavily reviewed practice could in principle be missed. The claim “most reviewed practice” means most reviewed in this sample, drawn as described.

We sell software that measures and improves AI visibility, which is why we had the tooling to run this. The data is what the engines returned; every question, every answer classification and every citation is downloadable below, and anyone can reproduce the checks by asking the same questions.

Download the raw data

Both files are CC-BY 4.0. Cite as “Sagelyn Research, September 2026”.

• Per-practice summary (CSV) — per practice: reviews, rating, then for each run the score, answers named, checks measured and own-site flag, plus both-run totals.
• Per-question detail (CSV) — all 1,440 checks across both runs: run, date, practice, engine, question, answer role, error if any, competitors named, every URL cited.

What people ask about this