15 questions patients ask AI before choosing a diagnostic lab, and what answers them
Before a patient opens a single diagnostic lab's website, they ask an AI assistant which lab is cheapest, does home collection, and is open right now, and the answer depends on facts the lab never wrote in a form a machine can read.
A patient does not start with a lab's website anymore. They open ChatGPT or Google's AI Overview and ask something like 'cheapest thyroid test near me' or 'does this lab pick up samples from home', and the AI answers from whatever text it can actually parse. If a lab's price list is a JPEG and its hours live inside a banner graphic, the AI has nothing to quote, so it quotes a competitor instead.
I scanned 18 real Indian diagnostic and pathology lab homepages the same way an AI crawler would, checking for the plain text and structured data these assistants read. The median AI readability score was 79 out of 100, ranging from 23 to 96. But the median hides the real story: the 11 labs with structured data (JSON-LD schema) scored a median of 83, and the 7 without it scored a median of 46. That is a 37 point gap, the clearest dividing line in the whole sample.
Only 3 of 18 labs have FAQPage schema, the format that lets an AI lift a direct answer like 'is fasting required' straight from the page. Only 5 of 18 have a single clear H1 heading, the most basic on page signal there is. Below are 15 real questions a patient or caregiver asks an AI before choosing a lab, and the specific fix that makes a lab's own site the one the AI cites.
The 15 questions, and the fix for each
List the actual test price in plain text next to the test name, and wrap it in a Service or Offer schema block with a price field, not inside a price list graphic. An AI reading a JPEG has no price to quote.
If the price is not text, it does not exist to an AI.
State 'home sample collection available' as plain visible text on the homepage and inside the site's LocalBusiness or MedicalBusiness schema, ideally as a listed Service with areaServed set. A banner image saying the same thing tells a human, not an AI.
A service is only real to an AI when it is written, not shown.
Write the accreditation and its certificate or registration number as plain text, and add it to the schema as a hasCredential or award property on the organization. Only 11 of 18 labs in my scan have any schema at all, so most accreditation claims currently live outside anything an AI can parse.
An accreditation nobody can machine read might as well be unaccredited.
State the exact turnaround (same day, 24 hours, 48 hours) in plain text next to the relevant test or package, not buried on a separate policy page. This is a factual property an AI can lift directly if it sits in readable text near the test name.
Vague turnaround language loses to a competitor's specific number.
Give the report portal a plain text label like 'download your report here' with a real link, and answer the question directly inside an FAQPage schema entry. Only 3 of 18 labs in my scan carry any FAQ schema at all.
A process only an app icon explains is invisible to a text based AI.
List every test included in the package as plain text bullet points with the total price, and mark the package as a Product or Offer in schema. A poster graphic showing the same list gives an AI nothing to extract.
Package contents belong in text, not in a flyer image.
State fasting requirements (12 hour fasting required, or no fasting needed) as plain text right next to each test name, and repeat the answer inside an FAQPage entry. This is exactly the kind of yes or no fact answer engines prefer to lift verbatim, and only 3 of 18 labs have the schema to make that easy.
Yes or no facts are the easiest wins, and most labs still skip them.
Add an OpeningHoursSpecification block to the site's schema and also state the hours in plain text on the homepage, not only inside an image banner. This is a single, well defined schema type built for exactly this question.
Hours belong in a schema field, not a graphic no crawler can read.
Give every branch its own address with latitude and longitude in a PostalAddress and geo schema block, ideally one LocalBusiness entry per branch rather than a single head office listing. Regional and local labs scored a median of just 49 in my scan, the lowest of any category, and a missing per branch address is a common reason why.
One address for ten branches means an AI can only ever find one of them.
Put a clear, single call to action in plain text (book a test) under one unambiguous H1 heading that names the lab and what the page does. Only 5 of the 18 labs I scanned have one clean H1, the most basic signal an AI uses to understand what a page is even for.
If an AI cannot tell what your page is about from the H1, it will not send anyone to book there.
State the home collection fee, or that it is free, in plain text at the point where the service is mentioned, not only inside a terms and conditions PDF. A hidden fee buried in a document an AI does not crawl reads to the assistant as no answer at all.
An unstated fee is not neutral, it is a missing answer that sends the patient elsewhere.
List individual tests by their common patient facing names, not only internal codes, on a plain text page or as a Product or Offer per test, so an AI matching the exact test name in the question can find and cite the match.
A test only exists to an AI if it is named the way a patient names it.
Publish concrete, checkable trust signals as plain text: accreditation numbers, years in operation, review counts, inside a LocalBusiness schema with a review or aggregateRating field where genuinely available, not just a marketing adjective like 'trusted'. Structured, checkable claims are what separated the median 83 scoring labs with schema from the median 46 scoring labs without it.
Trusted is not a fact an AI can verify. A number is.
State which insurers, TPAs, or corporate programs are accepted as plain text, and answer it directly as an FAQPage entry if the list is short. Right now 15 of 18 labs in my scan have no FAQ schema at all to hold an answer like this.
A yes or no about insurance is a five minute fix most labs have not made.
Comparison questions favor whichever lab has the most complete, checkable structured facts (schema, FAQ answers, a clean H1, real prices) rather than the most persuasive copy. Regional labs, the ones a nearby patient is actually asking about, scored a median of just 49 in my scan, so most are currently unequipped to win this exact comparison.
AI does not compare marketing claims, it compares whichever site gave it the clearest facts.
What I found scanning 18 diagnostic lab sites
The single biggest split in the data is structured data. 11 of the 18 labs carry JSON-LD schema (LocalBusiness, MedicalBusiness, or similar) and those labs have a median AI readability score of 83. The 7 labs without any schema sit at a median of 46. Schema does not just decorate a page, it hands an AI a labelled fact (this is our address, this is a service we offer) instead of forcing it to guess from a paragraph of marketing copy.
FAQ content is close to absent. Only 3 of 18 labs mark up their FAQ section with FAQPage schema, which means the other 15 have written answers to questions like 'do you offer home collection' somewhere on the page, but not in a format an AI can lift as a direct answer. And only 5 of 18 have one clean H1 heading naming what the page is about, so 13 labs are asking an AI to guess their own headline.
Split by category, national chains scored a median of 82, at home first digital labs scored 74.5, and regional or local labs scored 49, the lowest of the three. That last number matters most for the question a nearby patient actually asks, a 'lab near me' search. The labs most likely to answer a hyper local, urgent query (a lab within driving distance, open today) are the least readable to the AI doing the answering.
The pattern under all 15
Across all 15 questions above, the pattern repeats. Labs publish their prices, hours, package details, and accreditation as banner images, PDF brochures, and poster graphics designed for a person scrolling on a phone, not for a language model reading HTML. A human squints at a price list JPEG and figures it out. An AI assistant sees an image with no readable text and moves on to the next result.
The fix is not a redesign. It is putting the same facts already on the page into plain text an AI can parse, and into the schema types (Service, Offer, FAQPage, LocalBusiness, OpeningHoursSpecification) built for exactly this purpose. A lab that does both stops competing on price alone and starts showing up as the answer. You can grade your own lab site free to see which of these it already passes.
Frequently asked questions
Why don't diagnostic labs show up when I ask ChatGPT for a lab near me
Most labs publish their prices, hours, and services as images or PDFs, which an AI assistant cannot read as text. In my scan of 18 labs, the 7 with no structured data at all scored a median AI readability of just 46 out of 100, so the AI has almost nothing reliable to cite.
Does adding schema markup actually help a lab get cited by AI
In my scan, the 11 labs with JSON-LD schema scored a median of 83, compared with 46 for the 7 without it. That 37 point gap was the clearest pattern in the whole sample, so yes, schema is doing real work.
What is the single most common gap on diagnostic lab websites
Two basics are both missing on most sites: only 5 of the 18 labs have one clean H1 heading, and only 3 of 18 mark up their FAQ content with FAQPage schema. Both are simple to fix and both are what an AI looks for first.
Do patients really ask AI assistants before choosing a lab
Yes. Questions like cheapest thyroid test near me, is this lab NABL accredited, and does this lab do home collection are exactly the kind of practical, comparison shopping queries people now type into ChatGPT or Google's AI Overview before opening any lab's website.