CLICK CODED · Never Not Working
People already ask ChatGPT and other AI assistants real health questions. Verified live today, not pulled from an old dataset: two of the most trusted hospital brands in the country return a hard HTTP 403 to a plainly-identified AI fetcher — including OpenAI's own GPTBot user-agent, the exact one their own published crawler-access documentation tells webmasters to look for. When an assistant tries to read either site directly, it gets nothing. It answers from somewhere else instead.
This isn't necessarily a robots.txt rule written specifically to target AI — neither site's robots.txt even mentions GPTBot or the other named AI crawlers by name. It's almost certainly a bot-protection service (a WAF or CDN-level filter) rejecting anything that doesn't look like a real browser, which incidentally blocks the same respectful, identified AI crawlers that OpenAI, Anthropic, and Perplexity explicitly ask sites to allow. The effect is the same either way: an assistant trying to read the page in real time gets a wall, not an answer.
This is 2 of 14 major U.S. hospital systems audited for The State of AI Visibility 2026 — Healthcare Systems Edition, this studio's own real benchmark. In that same audit, Kaiser Permanente and Cedars-Sinai also failed to return a usable homepage under the benchmark's own methodology and were excluded from scoring rather than counted as a 0 — the same honest-scope rule used across every benchmark this site publishes. Cleveland Clinic, the one system that let the fetcher straight in, was also the only one in the entire 14-system sample with a published llms.txt, and led the reachable field at 90/100.
| Hospital system | What happened |
|---|---|
| Mayo Clinic | HTTP 403 to an identified fetcher — re-verified live today |
| Johns Hopkins Medicine | HTTP 403 to an identified fetcher — re-verified live today |
| Kaiser Permanente | Failed to return a usable homepage, per the healthcare benchmark's own dated methodology |
| Cedars-Sinai | Failed to return a usable homepage, per the healthcare benchmark's own dated methodology |
| Cleveland Clinic | Reachable, scored 90/100, the only system with a published llms.txt |
Nobody is claiming an AI assistant should replace a doctor. The real issue is narrower and more mechanical: when someone asks an AI a real question about, say, Mayo Clinic's own actual guidance on a condition, and Mayo's own site won't let the assistant in, the assistant doesn't just say nothing — it answers from whatever source it can reach, which may be a forum post, a less careful competitor, or a stale cache. The institution with the most credibility on the exact question being asked is, mechanically, the one least likely to be the source of the answer.
Free, 30 seconds, no signup: the AI visibility checker. Full healthcare industry data: the complete 14-system benchmark.
Methodology: a real, direct server-side fetch using a clearly-identifying User-Agent, re-run and re-verified at time of publication for the two headline claims above. Broader 14-system data sourced from this studio's own dated, published healthcare benchmark, not re-derived independently for this page. Built by Click Coded, AI-operated and human-reviewed.