Why AI search ignores most dental practices

In a June 2026 scan of 527 dental practice websites across seven US metros, 96.2% (507 of 527) had no VISIBLE record in either ChatGPT or Perplexity, including failed API requests: not a confirmed zero-mention rate. Separately, 96.7% of rated practices sit at 4.3 stars or higher (SOCi's 2026 average for ChatGPT-recommended locations), 93.4% had robots.txt open to tested AI crawlers, and 53.9% had no detected LocalBusiness or Dentist markup. These whole-sample counts do not establish why a practice went unnamed.

527 dental practice websites were scanned across seven US metros: Austin, Charlotte, Columbus, Nashville, Raleigh, Salt Lake City and Tampa, plus nine smaller towns, in June 2026. Each site was checked for star rating on file, robots.txt access, and schema markup, then ChatGPT and Perplexity were asked directly whether they would name each practice.

96.7% of practices with a rating on file (506 of 523) already sit at or above 4.3 stars, the average SOCi's 2026 index found among the locations ChatGPT actually recommended, not a cutoff to clear. The median practice among those 523 carries 250 reviews and a 4.9 star average. Because nearly all of them are already there, the rating isn't what separates the named from the unnamed. Reviews are not the blocker.

93.4% of practices scanned (492 of 527) had robots.txt open to AI crawlers. Separately, a national study found 11.4% of dental sites blocked at least one of 13 tested AI and data crawlers, narrowing to 7.6% when limited to crawlers that OpenAI, Anthropic or Perplexity themselves operate. Either way, for the large majority, crawler access is not the blocker.

53.9% of the 527 practices scanned (284 of them) have no detected LocalBusiness or Dentist schema markup, a block of code that spells out the business type, location and treatments for a machine, rather than requiring an AI to reconstruct that from marketing copy. Only 3.4% (18 of 527) have both local-business and FAQPage markup detected. These are whole-sample counts, not a measured schema difference between named and unnamed practices.

ChatGPT named 3 of the 415 practices it could read for (0.7%). Perplexity named 18 of the 297 it could read for (6.1%). Three and eighteen names is not enough to build a formula from, which is why this research looks at the majority pattern instead of the small set of winners.

The whole-sample counts do not establish a schema difference between named and unnamed practices. We have not run a controlled test proving that adding schema markup gets a practice recommended. The honest version is still useful: if more than half of practices are missing basic machine-readable information about themselves, that's worth fixing regardless of whether it guarantees a citation tomorrow.

Start with readability, not reviews: can an AI crawler fetch the site, and can a machine parse what the practice is? The June 2026 scan of 527 dental practices measured these site properties, but its API failures prevent treating the unnamed group as confirmed zero mentions.

In a June 2026 scan of 527 US dental practices, 96.2% had no VISIBLE record in either ChatGPT or Perplexity, including failed API requests. This is not a confirmed zero-mention rate. Across the whole sample, 53.9% had no detected LocalBusiness or Dentist schema; that does not mean all structured data was absent or establish a cause of non-mention.

No. 96.7% of practices scanned already sit at 4.3 stars or higher, the average SOCi's 2026 index found among the locations ChatGPT actually recommended, not a cutoff to clear. Because nearly all of them are already there, review quality does not explain which practices get named and which don't.

For most practices, no. 93.4% of the practices scanned had robots.txt open to AI crawlers. Separately, a national study found 11.4% of dental sites blocked at least one of 13 tested AI and data crawlers, narrowing to 7.6% when limited to crawlers that OpenAI, Anthropic or Perplexity themselves operate. Crawler access is not the primary blocker for the large majority.

No. These whole-sample schema counts do not establish that practices missing from AI answers were disproportionately missing markup. They are not evidence that adding schema causes recommendations. Check business markup as infrastructure, without expecting a guaranteed citation.

ChatGPT returned a usable reading for 415 of the 527 practices; the rest hit a rate-limit error. Perplexity returned a usable reading for 297 of 527; the rest hit an authentication error, a different failure than ChatGPT's. Failed checks were excluded from each engine's usable-reading denominator, but remain in the whole-sample no-VISIBLE total. A failed request is not evidence of invisibility.

Strong. Among the 523 practices with a rating on file, the median practice carries 250 reviews and a 4.9 star average, and 96.7% already sit at 4.3 stars or higher, the level SOCi's 2026 index found among locations ChatGPT actually recommended. Reviews are close to a solved problem across this group, which is part of why we don't treat them as the blocker.

Unevenly, and for different reasons in each city. In Charlotte, 154 practices, ChatGPT named 2. In Columbus, 58 practices, ChatGPT named 0. In Austin, 111 practices, Perplexity returned a usable reading for all of them and named 7, but ChatGPT could only get a usable reading for 21 of those same 111. In Tampa, 42 practices, ChatGPT named 0 and Perplexity couldn't be read at all in that scan. That split shows two different problems hiding under one word, "invisible": being read and not picked, versus not being readable at all.

It's a block of code that spells out the basics of a business for a machine to read directly, rather than making it guess from marketing copy — the type of business, its location, the conditions it treats. Think of it as a label on a can instead of a description buried somewhere on the box: a machine doesn't have to reconstruct what's inside from prose.

ChatGPT named 3 of the 415 practices it could read for, 0.7%. Perplexity named 18 of the 297 it could read for, 6.1%. Three names and eighteen names isn't enough to build a reliable formula from, which is why this research focuses on what the majority that had no VISIBLE record have in common instead.

No, and the city-by-city results show why. One version is "we looked and didn't pick you" — the engine could read the site but chose not to name it. The other is "we couldn't even read you" — a crawler or access problem stopped the engine from forming an opinion at all. The second is usually the cheaper problem to fix, but the two need different fixes, so it's worth knowing which one applies to you before acting.

In two layers. First, the engine has to be able to read the site: 93.4% of the 527 practices scanned in June 2026 had robots.txt open to AI crawlers. Second, a machine has to be able to parse what the practice is: 53.9% of scanned practices had no detected LocalBusiness or Dentist schema markup. Clearing both layers still does not guarantee a mention. In that scan, ChatGPT named 3 of the 415 practices it could read for, and Perplexity named 18 of the 297 it could read for.

In the June 2026 scan, yes on rate: Perplexity named 18 of the 297 practices it could read for (6.1%), while ChatGPT named 3 of the 415 it could read for (0.7%). The two engines also failed differently: ChatGPT hit rate-limit errors on the checks it could not complete, Perplexity hit authentication errors. Both counts are small, which is why this research looks at the majority pattern instead of the small set of winners.

Yes. The narration is synthesized, but the data, methodology and limitations discussed in it are Citevio's own and are the same figures and sources linked throughout this page.