Lab Website Benchmark: 30 Sites, Read Like an AI Crawler

Why I ran this

A patient who needs an HbA1c test doesn’t search for a laboratory. They search for the test. Their doctor’s office searches for turnaround times and specimen requirements. And more of both now start by asking ChatGPT, Perplexity or Google’s AI Overview instead of scrolling a results page.

Diagnostics is one of the two sectors I focus on most (see SEO for pathology labs), and I wanted a baseline built from evidence rather than opinion. So I read 30 lab websites the way a crawler does.

How I checked each site

The sample: 30 patient-facing pathology and diagnostic lab websites across the US, UK, Australia and Canada. National chains, regional labs and direct-to-consumer testing brands. For each one I read three things:

1. The homepage HTML exactly as served.
2. The robots.txt file.
3. The XML sitemap.

No logins and no JavaScript rendering. That is deliberate: several AI crawlers fetch pages without running scripts, so the raw HTML is close to what they get. Google renders JavaScript, so a site can show Google more than it shows an AI crawler; I note that limit below.

Half the sites have no page for an individual test

15 of the 30 sitemaps contained no URL I could identify as a test page. Only 12 had ten or more. For a business whose customers search test by test, that is the single biggest gap in the sample.

Without a page for “thyroid panel” or “vitamin D test” there is nothing for Google to rank for that search and nothing for an AI answer to quote. The aggregator that does have the page gets the patient.

The labs that had built test pages had usually built hundreds. It is close to all or nothing, which tells me the gap is a decision that was never made rather than a budget that ran out.

Most labs don’t describe themselves to machines

13 of the 30 homepages carried no structured data at all. Only 7 used any medical schema type (MedicalOrganization, DiagnosticLab and similar), and just 3 used MedicalTest, the type built to describe a test.

Schema.org has had a DiagnosticLab type with an availableTest property for years. Almost nobody in this sample uses it.

Structured data won’t rank a thin page on its own. What it does is remove guesswork: what this organisation is, which tests it offers, where it collects samples. That is exactly the question an AI answer engine is trying to resolve before it names a lab. My medical schema markup guide shows how I build that graph.

The crawlers are welcome. The content isn’t ready for them.

Only 1 of the 30 sites blocked any of the major AI crawlers in robots.txt. So this isn’t a policy problem: labs have left the door open.

But 24 of the 30 had no llms.txt file, a plain-text summary some AI tools read to understand a site quickly. It is a small file and an early convention, not a ranking factor. Its absence is still a signal of who is paying attention.

The basics are still leaking

None of these is dramatic on its own. Together they describe homepages that were designed carefully and then never checked against how search engines read them.

What I would fix first on a lab website

1. Build a page for every test people search for. Start with the 30 to 50 most-ordered tests. Each page: what it measures, who it’s for, preparation, turnaround, price if you publish it, where to book. Group them by condition. This is the relevance signal both Google and AI answers are missing.
2. Give every collection site its own page and its own Google Business Profile. Hours, parking, walk-in or appointment, and the tests available there. Not the same template with the suburb swapped. I cover this in how pathology labs rank on Google Maps.
3. Describe the lab in schema. DiagnosticLab or MedicalOrganization on the homepage, MedicalTest on test pages, location data on every collection site.
4. Write for the doctor as well as the patient. A referrer section with the test menu, specimen requirements and courier details. Almost no lab in this sample does this well, and it is the side of lab search with the highest lifetime value.

Why this matters for agentic optimisation

AI agents are starting to act for patients: comparing providers, checking opening hours and starting bookings. An agent can only do that with facts it can read. A lab with no test pages, no structured hours and no booking link on each collection site gives an agent nothing to act on, so it moves to a lab that does. Preparing for that is what I call agentic optimisation, and for labs it starts with the same fixes listed above.

What this study doesn’t show

• It reads homepages, robots.txt and sitemaps only. A lab could have strong test pages that are missing from its sitemap, and the study would miss them.
• Test and location pages were identified from URL patterns, so the counts are estimates.
• Sites that render content with JavaScript may show more to Google (which renders) than to crawlers that don’t.
• 30 sites is a sample, not a census. I have not named any lab; the pattern is the point.

Want to see what your lab’s site shows?

I’ll run the same checks on your website and send you a one-page read of what search engines and AI crawlers can and can’t see, with the three fixes I would make first. It’s free, it’s private, and you’ll hear back within one working day. Request your lab’s snapshot.