ChatGPT Health relaunches to all US users

OpenAI began rolling out ChatGPT Health to all US users 18 and older on web and iOS on July 23, six months after a limited pilot that reporting described as drawing “lackluster” feedback. Users can connect Apple Health data and, through a new partnership with health-data aggregator b.well, electronic health records pulled from roughly 2.2 million US healthcare providers, letting ChatGPT reference a person’s actual conditions, medications, and history in conversation rather than answering generically. OpenAI said conversations inside ChatGPT Health will not be used to train its foundation models.

ChatGPT Health first launched to a small group of testers on January 7. The months between that pilot and this relaunch went into what reporting describes as a rebuild, with the record-linking integration as the headline addition this time around rather than changes to the underlying model.

The interesting part for builders isn’t the chatbot, it’s the plumbing: b.well’s aggregation layer is doing the hard, unglamorous work of normalizing records across thousands of providers so a general-purpose assistant can use them safely. That’s the same pattern playing out elsewhere in enterprise AI, where Claude’s new connector into Anthropic’s own usage data and similar data-access integrations matter more for what they unlock than for any change to the model itself. Expect more consumer AI health features to route through data aggregators like b.well rather than building record access from scratch, and expect the training-data promise here to face real scrutiny given how sensitive medical conversations are.

Update, September 1

OpenAI extended ChatGPT Health to the clinician side on September 1, integrating directly with Epic’s EHR system, which holds records for more than 325 million patients. This is a separate, organization-facing deployment for hospital systems and clinical teams rather than the consumer feature above: clinicians get read-only access to a patient’s appointment notes, lab results, medications, and specialist documentation, and can use ChatGPT to summarize a chart, prepare for a visit, or spot changes in a patient’s history without leaving it. OpenAI said the integration cannot write back to the record, and reported that internal testing across 4,300 physician responses spanning 27 clinical use cases came back 99.1% safe. The consumer- and clinician-facing products now form two distinct entry points into the same record-access plumbing, both routed through third-party health-data infrastructure rather than a model change.