Are Osaka's 400 aesthetic-related clinics readable by AI? A machine-readability survey

The precondition for being treated as a candidate in AI search is that the site is in a state where AI can read it. So what is the actual state on the medical-institution side? Following our previous study of the 67 sites on the explaining side, this time we conducted a machine-readability survey of the medical-institution side—400 aesthetic-related clinics in Osaka Prefecture.

Here is the design. From the medical clinics in Osaka Prefecture derived from the Ministry of Health, Labour and Welfare's Medical Information Network (i-Iryojoho), we took as our population the 1,171 facilities advertising any of cosmetic surgery, aesthetic dermatology, plastic surgery, or dermatology, and from those whose website URL could be confirmed, we extracted 400 facilities by a mechanical rule (no arbitrary selection). For each facility's site, we mechanically assessed the machine-readable items: the allow status for 12 major AI crawlers in robots.txt, the presence of llms.txt, sitemap.xml, structured data (JSON-LD), meta descriptions, and https (measurement date: July 13, 2026).

Here are the results. The number allowing all 12 major AI crawlers was 356 of the 400 facilities (89%). On the other hand, there were 8 facilities blocking GPTBot, ClaudeBot, and the like en masse. The number with llms.txt installed was 39 facilities (10%). Structured data (JSON-LD) output was present at 209 facilities (52%). sitemap.xml at 321 facilities (80%). The undeterminable rate was 8.9% (for sites rendered with JavaScript or platform-type sites, an assessment based on the initial HTML diverges from reality, so we distinguish these as undeterminable).

Placing these alongside the distribution of the 67 LLMO/GEO explainer sites surveyed last time (AI crawlers fully allowed 93%, llms.txt 27%, structured data 70%), the composition becomes visible. The allow rate for AI crawlers barely differs between the explaining side and the medical-institution side (93% vs. 89%). This is likely a reflection of the fact that many sites do not mention AI crawlers individually in robots.txt—remaining in their initial state. The gap widens for things that are set up actively: llms.txt was 27% vs. 10%, and structured data 70% vs. 52%. Note that the structured data on the medical-institution side likely includes what reservation systems and CMSs output by default, so the rate of intentional preparation may be lower than this figure.

How should we read this result? The vast majority of medical institutions are in a state of "not refusing AI, but not actively preparing for it either." The 8 cases of blanket blocking are each facility's legitimate choice, such as not wanting use for training. What matters is that getting the machine-readable form in order is by no means difficult. Checking robots.txt, outputting structured data, and installing llms.txt if needed—simply putting these in order places you, at least on the precondition of a "readable state," in the leading group of the current distribution. However, as this site's fixed-point observation shows, a readable state is merely the entrance to the conditions for being recommended. After getting readability in order, observe how your clinic is being treated on AI—that order is the realistic one.

Let us state the limitations explicitly. This survey is a mechanical assessment of the machine-readable form; it does not measure content quality or how sites are actually treated in AI search, nor does the presence or absence of these items indicate a facility's superiority or inferiority. The target is a single point in time in Osaka Prefecture (a nationwide survey is currently in progress and will be reported separately upon completion). (Added 2026-07-14: The nationwide survey has been published)