The rollout turns a walled-off pilot into an always-on option that can quietly inform ordinary conversations, expanding how much of Americans’ medical data lives inside an AI chatbot.
OpenAI began rolling out Health in ChatGPT to its general U.S. user base on July 23, 2026, letting logged-in adults on Free, Go, Plus, and Pro plans securely connect Apple Health and medical records from supported U.S. hospital systems, One Medical, and Function Health. Once connected, ChatGPT can draw on that history such as medications, lab results, recent visits, sleep and activity data across any conversation, not just inside a dedicated Health tab.
That’s the core change from OpenAI’s original health pilot, introduced earlier this year to a limited group of testers. In that version, users had to open a separate space to get the full benefit. OpenAI says early testing showed that setup didn’t match how people actually used the feature: more than 70% of health-related conversations happened outside the dedicated space, as users asked things like whether a recipe fit a dietary restriction or how to plan a lower-impact workout around a recent injury.
The context for the expansion is scale. OpenAI says more than 300 million people ask ChatGPT health-related questions every week. Folding personal medical context into that traffic, rather than keeping it siloed. It suggests OpenAI sees health as a default use case for the assistant, not a niche add-on.
The company is also pairing the rollout with model upgrades: GPT-5.5 Instant now handles health conversations for free users, while paid users get GPT-5.6 Sol, which OpenAI says outperformed its predecessor on HealthBench Professional, an evaluation built with input from physicians. OpenAI says doctors also stress-tested the connected-data experience before release.
On privacy, OpenAI’s stated safeguards are layered: connected medical records and Apple Health data get additional encryption, are excluded from model training and ad targeting regardless of a user’s other settings, and require permission prompts before ChatGPT uses them (which users can disable). Disconnecting a source deletes its synced data within 30 days.
