We asked 4 AI engines to recommend a longevity clinic. Here's how they decided.
We typed one question into ChatGPT, Claude, Gemini, and Google AI Mode: the exact question a prospective client types. Then we recorded everything each engine did before it answered. This is the full decision path, and it explains why some excellent clinics never appear in answers.
The question
[SCAN DATA: the exact Q1 query used, e.g. "best longevity clinic in [city] for a 45-year-old executive who wants a full health assessment". Use the real query verbatim.]
No brand names. No coaching. A fresh session on each platform, the way a real person arrives: with a need and no loyalty.
What the engines did before answering
None of the four answered from memory alone. Each ran searches, pulled sources, and composed. The interesting part is which sources.
[SCAN DATA: per platform, 2-3 sentences: number of sources consulted, the domains, which were editorial vs directories vs reviews. Include the capture-grid citation domains. One screenshot per platform.]
Across the four runs, [SCAN DATA: N] distinct sources were consulted. [SCAN DATA: N] appeared in more than one engine's pool. That overlap is the citation pool for this query, and it is smaller than most people expect: win those sources and you are structurally present everywhere; miss them and no amount of on-site polish rescues you.
Who got recommended, and why
[SCAN DATA: the recommended clinic(s), anonymized or named per client permission. The verbatim recommendation sentences from each engine.]
Read the engines' own phrasing closely and a pattern shows: every reason given traces back to something machine-readable. Services the engine could list because they existed as structured, plain-language data. Credentials it could verify because independent sources agreed. Prices it could state because they were published. The engines did not reward the best medicine. They rewarded the most legible medicine.
Who got skipped, and why that clinic never finds out
[SCAN DATA: the skipped clinic (anonymized), what the engines encountered: unparseable pages, zero citations, contradictory data. One damning specific, e.g. "its services page returned no extractable text at all."]
Here is the uncomfortable part. The skipped clinic was not rejected. It was never considered. There is no notification, no analytics event, no lost-deal report. The prospective client simply books elsewhere, and the clinic's dashboard shows another quiet day. Invisibility in AI answers is a silent failure mode, which is exactly why it persists.
What this means if you run a healthspan business
Three practical conclusions from the recordings. First, the citation pool for your money query is finite and discoverable: run the query, collect the sources, and you have your placement target list. Second, your own site is the evidence base engines quote from, so every claim should survive being extracted and repeated out of context. Third, the four layers we assess in every scan (technical, content, trust, action) are not a framework we invented; they are the four places we watched engines make their decision, in order.
We run this exact exercise for any healthspan business, free, with the recordings included. Run your scan.