What can search systems read on dental websites?

OSM nodes only; 131 dentist-tagged nodes, 31 with website/contact:website; excluded one Google Maps link, one dental imaging centre and one dental materials supplier. All remaining 28 distinct website hosts retained, including unreachable sites. No replacement during the four-week pilot. Bounds: 44.60, 20.20, 44.95, 20.75. This convenience sample is not representative of all dental practices.

We fetch robots.txt, the website entry page (including up to three redirects), llms.txt and the first declared sitemap or /sitemap.xml. We identify as AI1GEOObservatory, respect its robots rules, check websites sequentially, pause at least two seconds between requests within each website check and reject private network addresses. Unknown robots responses stop collection. A missing robots file (404/410) means no rule was found.

Responses are bounded to 1 MB (robots: 500 KB); truncated pages are not scored. Only HTML sent by the website is inspected: no JavaScript rendering, authenticated pages or third-party audit provider. Network errors, blocked requests and non-HTML responses are gaps, not negative findings.

JSON-LD “parseable” means valid JSON syntax, not complete schema validation. We inspect explicit schema.org Dentist/LocalBusiness nodes (including arrays and @graph) and report name, address, telephone and hours fields. Remote contexts, every subtype and client-rendered markup are outside this method. Canonical and noindex findings are observations, not proof of actual indexing.

Rules for Googlebot, Bingbot and OAI-SearchBot are evaluated for the final homepage path. GPTBot is recorded separately in CSV as a training control. We do not impersonate these crawlers or prove CDN access for their IP addresses. llms.txt is experimental and is not an AI ranking factor or required Google AI markup.

Method 1.1 (current): robots.txt groups are read per RFC 9309, so comments and blank lines no longer split a group. Noindex is read directive by directive: a directive scoped to another crawler (for example otherbot: noindex) and value directives such as max-image-preview:none no longer count. After the full pass, websites that failed with a network error or an unavailable robots.txt are retried once after a pause, as long as the run is within its 30-minute time budget; both attempts are recorded in the evidence JSON and the CSV attempts column, and a retry skipped for time is marked as such. Each issue states the method it was collected with; issues collected with method 1.0 had a single attempt and are not recomputed.

Coverage = successfully fetched, untruncated HTML homepages / the fixed sample. A new complete issue is published only at ≥80%. Each metric has its own observed denominator; missing observations remain unknown. Dated issues are immutable and use UTC dates. Weekly collection runs Monday at 08:00 Europe/Belgrade after the baseline.

AI-answer panel: NOT_RUN. Manual observations are planned separately for T0 and T+28. No AI mention counts or AI visibility scores are inferred from these technical signals.

Source retrieved: 2026-09-26. Data © OpenStreetMap contributors; measurements by AI1GEO. This derived database and CSV are available under ODbL 1.0. Reuse with attribution and the applicable share-alike terms.

Method 1.1 · Collection 2026-10-12T06:00:02.971Z · Collector 521e473f9fd3401212742d81f49aa23e7399491f · Cohort SHA-256 49b3bffbffb2c8b717293fc5fc000dbc97e9fd43c12a8b54f872f26ab9497770. Measurement evidence (JSON)