When someone asks an AI about a hospital group, what can it read?
Someone types the name of a pathology provider, a health fund or an aged care group into an AI tool and asks where to go or who to trust. What the tool can read on that organisation's homepage decides a lot about the answer. We looked at 91 large Australian health operators. We could read 70 homepages. On 59% of them, machines get a direct statement of who the organisation is. On 36%, they get nothing of the kind.
Who did we look at?
Hospital groups, pathology and imaging providers, pharmacy groups, health insurers, dental and fertility groups, aged care providers and large health charities. It is a mix, so the results describe that mix, not any one kind of provider.
What can a machine read on these homepages?
Of the 70 readable homepages:
- 41 (59%) carry JSON-LD structured data.
- 25 (36%) carry no structured data of any kind that we could see.
- 36 (51%) describe the organisation itself, and 28 (40%) describe the website.
- None uses FAQ markup. For sites that exist to answer questions from patients and families, that is the finding we found most striking.
The basics are mostly in place: 55 (79%) set a canonical address and 53 (76%) write a meta description. 42 (60%) set a social sharing image, and 38 (54%) have exactly one main heading.
What about the 12 sites built as JavaScript apps?
12 of the 91 show signs of being assembled in the browser by JavaScript, where the page we can read is only a shell. For those, the figures above probably understate what a visitor's browser actually sees.
Have these organisations taken a position on AI crawlers?
Hardly any. 68 of the 91 publish a robots.txt file, and 50 of those point to a sitemap. Fewer than 10 name an AI crawler, and fewer than 10 close the site to one. This is a choice about training and access, and we make no judgement on it.
11 of the 91 publish an llms.txt file. It is a proposal, not a standard, and no major AI service has said it relies on it.
What does a health organisation lose by leaving it to the machine?
We cannot measure a loss, and we do not claim one. What we can say is what is on the page: for an organisation where "who are they, where are they, what do they offer" is the question people ask, about a third of the readable homepages leave that question to be answered from page text alone. If your organisation's location pages, services and opening arrangements are described only in visible text, a machine has to infer them.
How is the next step different for a practice?
A large operator has a web team and a budget. A small practice usually has a template. In our companion study of small practices, 63% of readable homepages carried JSON-LD, close to the level here. Both groups share the same gap: the specific facts (services, location, how to book) have to be written by someone who knows them.
What can you do today?
Look at your own homepage the way a machine does. Our free AI Visibility Toolkit at /ai-toolkit reads it in about a minute and reports what is there. For any single page, such as a location or services page, the free AISearch Pulse extension for Chrome runs page checks, each tested against hand-checked web pages (November 2026), and gives a short fix for each one that fails, with up to 10 scans a day and no licence key. It shows no AI answers, and it is optimised for Australian businesses.
What this study does not cover
It says nothing about care or quality. It is not an audit. It reads homepages on one day. The mix of provider types is one hand-built list that is not a ranking and is not published. The method, the terms and the limits shared by all our scans are in How AISearch Global runs a technical-setup scan. We could not read 21 of the 91 homepages, which tells us about our scanner, not about those organisations, so every homepage percentage above is out of the 70 we could read.
Frequently asked questions
Why does FAQ markup matter for health sites?
FAQ markup is code that labels a question and its answer so a machine can pair them. Health sites are full of questions. None of the readable homepages we checked used it. We only looked at homepages, so inner pages may differ.
Should a health organisation block AI crawlers?
We make no judgement on it. It is a decision about training and access.
Does a missing structured-data block mean a hospital is hard to find?
No. People still find organisations through search, referrals and maps. This study looks at one input a machine can use.
Why group such different providers together?
This is a first look at the larger end of Australian health websites. Groups under 10 are suppressed, so the mix is reported as one set.
Sources
- AISearch Global health-operators snapshot, 22 November 2026 (private evidence held; aggregate counts only published).
- AISearch Global small-practices pilot, 22 November 2026.
- llmstxt.org, "The /llms.txt file" (page opened 22 Nov 2026; presents llms.txt as a proposal).
Viveka Mohan Das
Answer Engine Optimisation (AEO) and GEO Strategist, Published Author & Founder, AISearch Global. More about Viveka
Check your own homepage
The free AI Visibility Toolkit reads your homepage the way a machine does and reports what is there.