DEEP OCEAN RESEARCH №001

Can AI engines read Switzerland's private clinics?

Patients have started asking ChatGPT, Claude, and Perplexity for a clinic before they ever open a search page. So we scanned 24 Swiss private clinics to answer one question underneath that shift: can the machines even read them? 29% returned a non-200 response to at least one AI crawler user-agent; most of the rest are reachable but expose little structured data to parse. The two are different problems, and only the first is an access block.

24 Swiss private clinics scanned 7 crawlers tested 4 Jul 2026 · Switzerland Aesthetics · dental · fertility · dermatology

Archived 2026-07-11. Reports 001 to 003 are the readability series from an earlier version of our method. They stay online unchanged as part of the record, but they are delisted from the index. One claim below has been corrected: llms.txt and machine-readable briefs are hygiene; there is no evidence they cause AI citations.

Our current field sample runs the same test on a named category: see the US software field sample and our own dated record on The AI Citation Index. The other archived report in this series covers Swiss law firms and fiduciaries. Run this check on your own brand with the Sprint, $490.

KEY FINDINGS
88%
21 of 24

publish no llms.txt and no machine-readable brief. That is a hygiene gap, and public evidence shows these files do not cause AI citations on their own.

62%
15 of 24

expose no business entity an AI engine can parse to know who the clinic is.

54%
13 of 24

have at least one critical gap: AI cannot reach them, or there is nothing structured to read.

29%
7 of 24

returned a non-200 response to at least one AI crawler user-agent while serving Google normally, in a single test from our location.

THE READABILITY GAP

Five ways to be invisible

AI visibility starts with readability. Before an engine can recommend a business, it has to reach the site and parse who the business is. Across 24 sites, here is how often each layer was missing.

Share of Swiss private clinics with each gap

Percentage of 24 Swiss private clinics scanned, 4 Jul 2026.
No llms.txt brief88% · 21/24
No machine-readable business entity62% · 15/24
At least one critical gap54% · 13/24
Zero structured data at all46% · 11/24
Blocks at least one AI crawler29% · 7/24
CRAWLER ACCESS

Claude gets turned away most

Each site was requested with 7 crawler user-agents. These are the AI engines specifically, showing how often the server returned anything other than a normal 200 response from our test vantage.

Share of Swiss private clinics that blocked each AI crawler

Non-200 response to the engine's user-agent · 24 Swiss private clinics.
Claude · ClaudeBot29% · 7/24
ChatGPT · GPTBot17% · 4/24
ChatGPT Search · OAI-SearchBot17% · 4/24
Perplexity · PerplexityBot17% · 4/24
Google AI · Google-Extended17% · 4/24
3 of 24 sites (12%) returned nothing to any crawler at all, including Google's, which points to wholesale bot protection rather than an AI-specific choice. The rest of the gap comes from sites that returned a non-200 response to one AI crawler user-agent specifically while answering the others.

Methodology

SAMPLE
24 private, cash-pay clinics in Switzerland (aesthetics, dental, fertility, dermatology) across Zurich, Geneva, Lausanne, Lugano and St. Gallen. Hospital groups and directories excluded. Clinic names are withheld; only aggregate figures are reported.
WHAT WE TESTED
Per-engine crawler access (7 user-agents), presence of llms.txt, robots.txt and sitemap.xml, and homepage structured data (JSON-LD types, including whether a business entity is declared).
CRAWLERS
GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended, plus Googlebot and Bingbot as reference.
HOW
Every measurement is a single HTTPS request, and the report lists the exact curl command behind each one so a business's own web team can re-run it. A later run can return a different response; server behavior varies with IP, location, time, CDN, and WAF rules.
DATE
4 Jul 2026, single run.

What this report does not claim

  • This measures readability (whether AI engines can access and parse a site) not citation frequency. How often each business is actually named in AI answers would require a separate, prompt-based study.
  • Access was tested from a single vantage on one date. A few non-responses may reflect transient timeouts or geographic bot rules rather than a permanent block; findings are framed as "from our test vantage" for that reason.
  • Structured-data checks read the homepage only. An entity declared on a deeper page would not be counted here.