Three Platforms. Three Strategies. One Prize.

Well… I covered ChatGPT Health in January. Then Anthropic dropped their move.

And now Perplexity AI.

Ten weeks. Three major platforms. Same territory. And now the picture is starting to get clear enough to say something useful about it.

I’ve been watching where AI agent layers are heading across industries for a while now. Healthcare is where it gets personal. Literally. And what just happened in ten weeks is not a feature race in my humble opinion. It’s a positioning war for the layer that sits above everything else in healthcare – where you actually ask questions, make decisions, and route your health through your life.

The EHR stays the system of record. Nobody’s touching that.

What’s being contested right now is the system of engagement. And each player picked a different door. Sorta reminds me of “The Monty Hall Show”

OpenAI: The Engagement Flywheel

They didn’t build ChatGPT Health because they had a healthcare vision.

They built it because their users were already asking 40 million health questions a day inside a general-purpose chatbot. They looked at their own traffic and formalized what was already happening.

ChatGPT Health is the consumer front door. OpenAI for Healthcare – HIPAA-configured, already running at Cedars-Sinai and HCA Healthcare – is the actual bet.

The theory: consumer habit creates bottom-up institutional demand. Clinicians use it personally. Then they want it inside clinical workflows. The enterprise contract follows the habit.

Smart. Real distribution advantage.

Real problem: their own accuracy data is damaging. Studies are showing ChatGPT Health incorrectly triages more than half of medical emergencies. You can’t paper over that with a disclaimer. Health systems have legal teams. That gap doesn’t just slow enterprise sales – it becomes the headline when the first serious incident happens.

Distribution without clinical trust isn’t a moat. It’s a trigger event waiting for a date.

Anthropic: The Workflow Wedge

Anthropic’s own leadership said it plainly. Enterprise is the focus. Health systems, payers, pharma. That’s where the energy is going.

Claude for Healthcare targets the administrative backlog. CMS coverage policy. ICD-10. NPI registry. Prior auth. FHIR workflows. Clinical trial operations.

Unglamorous work. Prior auth alone costs US healthcare an estimated $31 billion annually in administrative friction. The ROI is immediate. The business case writes itself.

That’s actually how large healthcare organizations adopt technology. Not through consumer-facing demos. Through quantifiable friction reduction inside systems they already run.

The timing risk is real though.

Consumer habits are forming now. OpenAI is inside those habits. When a health system executive asks IT for “the AI our clinicians are already using at home” – and that question is coming – Anthropic needs to have already won the workflow layer before it gets asked.

If they haven’t, they’re fighting a brand recognition war.

Perplexity: The Personal Health Operating Layer

This one I believe could be underestimated. Significantly.

Perplexity Health isn’t a chat interface bolted onto health data. It’s built on Perplexity Computer – their agentic orchestration platform. EHR access through b.well’s network covering 2.4 million providers and 350+ health plans. Apple Health, Fitbit (now part of Google), Ultrahuman, Withings, OURA incoming. Purpose-built agents for nutrition, sleep, activity. A unified dashboard above every fragmented portal you currently navigate separately.

Your health data right now is scattered across many portals that don’t necessarily talk to each other. Labs in one place. Prescriptions in another. Wearables somewhere else. Insurance in a fourth. Fitness in a fifth.

Perplexity is betting they can sit above all of it. One interface. All the context at once. Grounded in your actual clinical history – not generic population data.

If that works – you stop going to MyChart separately – you know the one – 1970’s sort of looking thing. You stop logging into your lab portal. You stop opening your insurance site. You just ask.

That’s not a health app. That’s a fundamentally different kind of problem for everyone else in this space.

Their vulnerability is distribution. OpenAI has 800 million weekly users forming health AI habits right now. Perplexity has a smaller, highly engaged base. Being better isn’t enough. They need to be meaningfully better to overcome that gap.

Two Players I’m Deliberately Setting Aside

Microsoft and Google both belong in this analysis. Each needs separate treatment.

Microsoft may already be the infrastructure everyone else is building on top of. Epic integration. Azure Health Data Services. Teams embedded inside clinical workflows. Copilot Health launched March 12. They didn’t need a consumer product announcement to have a healthcare position. They’re already inside the workflow at scale.

Google has the most powerful underlying health data assets of anyone on this list – and a decade-long execution problem in regulated markets. Real capability. Inconsistent compounding into market position. MedGemma is technically impressive. Their healthcare history is a more complicated story.

Both deserve their own analysis. Both are watching this week closely.

What I’m Most Interested In

First order is obvious.

Consumer health chat is commoditized. Digital health products built on conversation plus prompts sitting on someone else’s data got repriced this week. If your defensibility was the chat layer – think hard.

Second order is what I think matters more.

The platform that assembles a unified personal health agent – labs, wearables, clinical history, medications, synthesized in real time – can own the most intimate high-frequency data relationship possible in a person’s life.

Health isn’t one vertical among many. It contextualizes every other decision you make. Financial. Professional. Personal. If an AI agent already knows your work, your communications, your finances, and now your complete health picture – you’ve crossed into something different. That’s not a healthcare story anymore. That’s a total information architecture story.

And the trust question underneath all three of these launches hasn’t been answered by any of them.

Every platform made the same privacy pledge this week. Encrypted. Not used for training. User controlled.

None of it has been independently verified – FYI

The legal architecture around consumer health data flowing through an AI intermediary is genuinely unsettled territory. HIPAA protects data at the covered entity. It doesn’t automatically follow data a consumer voluntarily routes through a tech platform. (Be warned)

What happens when it’s subpoenaed?

What happens if it surfaces in an insurer’s risk model through a data partnership buried in a terms update?

These are not hypothetical.

The platform that solves the trust architecture problem first – not just the privacy pledge, the actual architecture – controls the enterprise narrative when that moment hits.

It will hit.

My signature line for a while now has been: own the default, own the data; own the data, own the decade.

Healthcare is one of the most important defaults anyone in AI is racing to own.

The question is whether the three companies in this race understand that fully – or whether they think they’re building health products.