Case Study · Healthcare
AI knew this hospital. It wouldn't recommend them.
One of India's largest hospital networks had strong brand recognition with every major AI platform. They still weren't appearing when patients asked for help.
By Aditya Nittur · May 2026 · 6 min read
Cofounder at FlowBlinq. Built data-driven trading systems across five international exchanges. Brings quantitative rigour to FlowBlinq's AI commerce intelligence platform.
Key takeaways
- 1. Healthcare is YMYL — AI won't recommend a hospital it recognises but can't verify
- 2. A multi-city network had zero geographic signals — AI couldn't map them to any location
- 3. The fix was structural, not creative — schema, entity definitions, and llms.txt
- 4. Integration was three changes to existing files — no rebuild, no redesign
- 5. AI platforms now actively cite them across medical specialties and city-level queries
Recognition is not recommendation. That distinction matters more in healthcare than anywhere else — and it's exactly the problem this engagement uncovered.
When we ran the initial GEO audit on one of India's largest multi-specialty hospital networks, the results were split in an unusual way. AI platforms knew the brand well. They could describe it, categorise it, place it in the market. But when patients asked medical questions — which hospital for a specific condition, which city had a particular specialty — this network wasn't appearing.
Brand knowledge without citation is visibility without reach. And in healthcare, the queries that matter most are exactly the ones where trust has to be earned structurally.
The YMYL problem
Healthcare content sits in a category AI researchers call YMYL — “Your Money or Your Life.” When a patient asks ChatGPT or Perplexity which hospital to visit for cardiac surgery, the AI is making a recommendation that has direct health consequences. These platforms apply a higher bar before surfacing any result.
That bar is not about content quality. It's about verifiability. AI systems need structured proof — not just good copy — before they'll recommend a medical provider. A hospital with beautiful website content and zero schema markup will lose to a smaller clinic that has machine-readable credentials, structured service definitions, and proper location data.
This network had the content. They didn't have the structure.
What the audit found
Three gaps stood out as most responsible for the disconnect between recognition and recommendation.
No geographic signals
A hospital network operating across multiple cities had no location-level structured data. AI platforms had no way to answer “best cardiac hospital in Bangalore” or “orthopaedic specialists in Hyderabad” with this network as the answer — because nothing in the site's machine-readable layer connected specialties to locations. Every city-level query was effectively invisible to them.
No entity definitions for medical services
The site described procedures and departments in plain language written for patients. That's good for humans. AI systems, however, need to parse the relationship between a medical entity (a condition, a procedure, a specialty) and a provider. Without structured entity definitions — using vocabulary AI systems actually recognise — the content existed but wasn't usable for recommendation.
Structured data was nearly absent
Across thousands of pages covering departments, doctors, and procedures, there was almost no JSON-LD schema markup. AI crawlers visiting the site had no machine-readable signal about what was being offered, who was offering it, or where. The content was rich; the signal was absent.
What we proposed
The recommendations focused entirely on making existing content machine-readable — not on creating new content or redesigning anything.
Deploy schema markup at scale
Add JSON-LD schema to every department page, doctor profile, and procedure description. Use MedicalOrganization, Physician, and MedicalProcedure vocabulary. This gives AI systems structured facts to parse rather than forcing them to interpret unstructured HTML.
Define entities for every specialty and location pairing
Build structured entity definitions connecting each medical specialty to the specific locations where it's available. This directly addresses the city-level query gap. Once AI platforms can map “cardiology” to “Mumbai facility,” they can answer location-specific medical queries accurately.
Publish llms.txt and structured FAQ content
Create an llms.txt file that describes the site's structure to AI crawlers — what specialties exist, how locations are organised, where to find specific information. Supplement with FAQ markup covering the most common patient questions across each specialty. These give AI platforms extractable content for direct citation.
Structure doctor authority signals
YMYL content is evaluated heavily on author expertise. Doctor credentials, qualifications, and clinical experience need to be in structured markup — not just in readable bios. Adding Physician schema with qualification and affiliation fields directly addresses the trust gap AI platforms apply to medical recommendations.
The integration footprint
FlowBlinq's GEO layer deployed through three changes to the site's existing files — a beacon tag, a schema-injection block, and three rewrite rules serving the machine-readable files AI crawlers consume. No rebuild. No CMS migration. The integration runs on Appiness's existing Apache infrastructure and is fully removable.
What changed
The most immediate shift was in geographic visibility. Before, city-level medical queries returned no results from this network. After deploying location-specialty entity definitions, AI platforms began surfacing them for location-specific queries — the highest-intent category for healthcare search.
Structured data deployment moved AI platforms from “aware of this brand” to “able to recommend this brand.” The gap between those two states is what GEO closes.
Traffic to the site increased materially in the weeks following activation — driven by new AI-referred visitors arriving from medical queries rather than traditional search. The managing agency independently validated that the traffic increase was attributable to real human visitors, not crawler activity.
Still in progress
Competitive positioning — how AI platforms compare this network to peer institutions — remains the hardest pillar to close in healthcare. Clinical outcome data and peer-reviewed evidence, when structured for AI extraction, are what move this score. That work is ongoing.
The broader point
Healthcare is the most consequential vertical for GEO because the stakes of an AI recommendation are highest. But the mechanics are the same as every other vertical: AI platforms recommend what they can verify, not just what they know.
Every hospital, clinic, and health system that has invested in content marketing is sitting on the same gap. The content is there. The structure that turns content into citations isn't.
Find out whether AI platforms can verify what you offer.
Run a free AI visibility audit