AI Health Tools Are Proliferating Faster Than Evidence They Actually Work
Microsoft launched Copilot Health earlier this month as a dedicated space within its Copilot app, allowing users to connect medical records and query their health data directly.
4. AI Health Tools Are Proliferating Faster Than Evidence They Actually Work
Microsoft launched Copilot Health earlier this month as a dedicated space within its Copilot app, allowing users to connect medical records and query their health data directly. Days before that, Amazon expanded Health AI, an LLM-based tool previously available only to One Medical members, to a broader user base. The back-to-back announcements signal a coordinated push by two of the largest cloud and consumer technology companies to own a slice of AI-assisted personal health, a market where Google has also been active through its Med-PaLM and health-adjacent Gemini integrations.
The competitive dynamic here is less about clinical outcomes and more about data gravity. Microsoft gains access to structured medical record data through Copilot Health integrations, feeding a feedback loop that strengthens its health AI models and entrenches Azure as the backend of choice for healthcare systems. Amazon's move to open Health AI beyond One Medical's subscription wall is a land-grab for consumer health behavior data at scale, with the added effect of making One Medical's premium positioning harder to justify. The losers in the near term are standalone health AI startups like Babylon Health successors and symptom-checker platforms that cannot match the distribution leverage of Microsoft's 1 billion Windows users or Amazon's Prime ecosystem. Patients and clinicians, meanwhile, face a product landscape where deployment speed is clearly outrunning clinical validation.
The deeper structural signal is that Big Tech is reframing AI health tools as consumer features rather than regulated medical devices, a deliberate positioning choice that sidesteps the FDA's Software as a Medical Device framework and the slower evidence cycles of peer-reviewed clinical validation. If that framing holds, the competitive moat shifts entirely to distribution and trust, not efficacy, which is a troubling precedent for a domain where the cost of a confident wrong answer is materially different from a bad restaurant recommendation.