ChatGPT's Health Feature Is a Bet That the Value Was Always in Aggregation

OpenAI rolled out “Health in ChatGPT” to all logged-in US users 18 and up across Free, Go, Plus, and Pro, on web and iOS. Users can connect Apple Health, supported medical records, One Medical, or Function Health so ChatGPT can see labs, medications, activity, and sleep data in one place, compare new results against prior tests, and summarize what’s changed since a past appointment. OpenAI’s stated reason for building this now: over 300 million people already ask ChatGPT health questions weekly, and more than 70% of health conversations among early testers happened outside any dedicated health space — the demand for a health-aware assistant already existed inside general chat, unaggregated. The company says connected health data is never used for model training or ad targeting, layered with purpose-built encryption and isolation, and users control when it’s accessible; a memory feature carries insights forward across conversations. OpenAI is explicit that the feature supports care rather than replacing it — no diagnosis, no treatment.

The bet underneath the feature is that most of the value in health AI was sitting in the aggregation gap — one person’s records spread across five systems that don’t talk to each other. Andreessen Horowitz made a version of the same bet a month earlier, leading a $30 million Series A into Prosper AI, which automates the voice-heavy administrative work behind clinic operations: scheduling, insurance eligibility, billing. Prosper’s pitch is explicitly built for combinatorial complexity — specialty practices with 10-plus locations, 100-plus appointment types, 1,000-plus insurance plan combinations — and it escalates to human staff only when it can’t finish a task itself.

Read together, ChatGPT Health and Prosper are the same wager pointed in opposite directions: one aggregates a patient’s fragmented records so the patient can understand them, the other aggregates a clinic’s fragmented back-office systems so staff don’t have to reconcile them by hand. For consulting clients in health systems and payer organizations, both point to the same near-term governance question — not whether to adopt AI in health workflows, but who owns the data-unification layer once patient-facing and staff-facing AI are both reaching into the same fragmented records.