Verified nationwide across 7,643 facilities — are aesthetic clinic websites in a state AI can read?

When we recently published the machine-readability survey (a survey of machine readability — whether a website is in a state that AI can read) of 400 aesthetic-related clinics in Osaka Prefecture, we wrote that "the nationwide survey is currently in progress." That nationwide survey is now complete, so we report it here as a follow-up.

The design. Among the medical clinics nationwide derived from the Ministry of Health, Labour and Welfare's Medical Information Network (i-Iryojoho), we took as the population the 11,421 facilities advertising any of cosmetic surgery, aesthetic dermatology, plastic surgery, or dermatology, and surveyed all 7,643 facilities whose website URLs could be confirmed. This time it is not sampling but a complete census. The judgment items are identical to the Osaka survey: we performed machine judgment of external items such as the permission status of the 12 major AI crawlers in robots.txt, the installation of llms.txt, sitemap.xml, structured data (JSON-LD), and https (measurement period: July 13–14, 2026. Access to each facility's site was made only once, at spaced intervals).

The results. Of the 7,643 facilities, 6,776 (88.7%) allowed all 12 major AI crawlers. Those blocking some or all were 172 facilities (2.3%). Those with llms.txt installed were 697 facilities (9.1%). Structured-data (JSON-LD) output was found at 3,060 facilities (40.0%).

Placed alongside the distribution of the earlier 400 facilities in Osaka (fully allowed 89%, llms.txt 10%, structured data 52%), the AI-crawler allow rate and llms.txt almost matched. The Osaka figures were not region-specific but can be said to have represented the industry-wide present state. Only structured data came out 12 points lower nationwide, but this may be due to the difference in extraction method (the Osaka survey extracted 400 facilities, prioritizing the advertising of cosmetic surgery and aesthetic dermatology, whereas the nationwide census includes dermatology), and we do not assert it.

There is one finding that became visible for the first time in this complete census. Of the 7,643 facilities whose URLs are registered in public databases, only 6,832 (89.4%) could have their top page reached normally. The remaining 811 facilities (about one-tenth) could not have the site itself reached: the domain could not be resolved (358 cases), the page did not exist (404 in 214 cases), or there were timeouts, SSL errors, and the like. Closure, relocation, domain expiration, URL changes due to a redesign — the reasons are thought to be various, but what they have in common is that "the public registration information and the actual state of the web are out of sync." When AI talks about medical institutions, such obsolete information can also become part of the material. Confirming where and how your own clinic's information is registered has meaning as a basic step even before AI countermeasures.

A note on how to read this. This survey is a machine judgment of external form; it does not measure the quality of content or how it is actually treated in AI search, nor does the presence or absence of an item indicate the superiority or inferiority of a facility. For sites rendered with JavaScript or platform-type sites, the judgment may diverge from reality, and these are distinguished as unjudgeable (the unjudgeable rate is 7–11% depending on the item).

The picture in which the vast majority of medical institutions are in a state of "not refusing AI, but not actively preparing for it either" has not changed since the Osaka survey. Preparing the external form is not difficult, and once it is prepared, in terms of preconditions you join the leading group. What comes after that — how a site that has become readable is actually treated by AI — is the domain this site handles through actual measurement.