OpenAI is making ChatGPT Health available to every US-based user over 18 across all plans this week, expanding a feature that already fields 300 million health-related queries per week — up from 230 million in January 2026. The rollout lands one day after a Florida-based pastor sued the company over a near-fatal suggestion not to consult a doctor, giving the announcement a difficult backdrop. The feature moves out of a dedicated hub and into general chat, letting users pull from connected health data anywhere in the product.
The scale of health use on ChatGPT has grown faster than the hub could contain it. During testing, OpenAI observed that 70% of health-related queries took place outside the dedicated hub — users were asking medical questions in their normal chats regardless of where the company wanted to route them. The redesign concedes the point and lets connected health data inform responses across the full product surface.
Users can integrate personal data from Apple Health, Function, and MyFitnessPal, and pull medical records from hospital systems including Epic and Oracle Health, plus platforms like One Medical and Function Health. A user asking about a meal or an allergy in a general chat can now draw on that connected information without switching contexts. OpenAI said it does not use this data to train its models and works with physicians to improve health responses.
Key facts
- 01ChatGPT Health rolls out this week to all US users 18+ on Free, Go, Plus, and Pro plans across web and iOS.
- 02Weekly health-related queries on ChatGPT have climbed to 300 million, up from 230 million in January 2026.
- 0370% of health-related queries during testing happened outside the dedicated health hub, prompting the general-chat expansion.
- 04The launch comes one day after a Florida pastor sued OpenAI over a near-fatal suggestion not to consult a doctor.
- 05GPT-5.6-Luna, the smallest model in OpenAI's latest release, outperforms GPT-5.5 on the company's HealthBench evaluation.
The model side has kept pace. OpenAI said GPT-5.6-Luna, the smallest model in its latest release, outperforms GPT-5.5 on HealthBench, the open-source benchmark the company built to evaluate large language models on health queries. That framing — a smaller model beating a larger predecessor on a domain benchmark — is the pitch OpenAI wants to make about safety-through-capability rather than safety-through-restriction.
The legal picture is harder to spin. The Florida lawsuit filed this week alleges ChatGPT told a user not to seek medical care, with near-fatal consequences. OpenAI pointed to its terms of service in response, noting the product is "not intended for use in the diagnosis or treatment of any health condition," and told The New York Times it is working on making health- and medicine-related answers safer. Whether a disclaimer buried in terms holds up against 300 million weekly health queries is now a question for a court.
OpenAI's messaging with the rollout leans on the disclaimer while quietly acknowledging the reality that hundreds of millions of people are already using the product this way. The company said it wants users to verify information and make medical decisions based on professional advice. That is the guardrail, and it is a soft one — the same guardrail that failed in the case now in court.
Competitors are not sitting out. Anthropic and Google have both launched health-oriented AI features in recent quarters despite a growing body of studies flagging reliability problems with medical advice from language models. The commercial pull is obvious: health is one of the most-searched query categories on the internet, and every frontier lab wants that traffic flowing through its assistant rather than a search engine.
The integrations with Epic and Oracle Health are the more consequential piece for the broader healthcare stack. Hospital record systems have historically been walled gardens, and getting patient-level data flowing into a consumer chatbot — with user consent — changes what a general-purpose assistant can plausibly do in a medical context. It also puts OpenAI in closer proximity to HIPAA-adjacent workflows without the company positioning itself as a covered entity.
The counterweight is that AI chatbots remain unreliable clinicians. Independent studies have repeatedly shown language models produce confidently wrong medical answers, and the Florida suit is a concrete instance of the failure mode critics have been describing for two years. HealthBench improvements narrow the gap but do not close it, and OpenAI's own terms acknowledge the product is not a diagnostic tool. The gap between what the terms say and what 300 million weekly queries suggest users actually do is the exposure.
For the AI market, the rollout signals that OpenAI has decided the growth from vertical health integration outweighs the legal and reputational risk of edge-case failures at scale. That is a bet other labs will now be pressured to match, and it locks in a pattern where consumer AI products absorb roles — first-line health triage, medication questions, symptom checking — that were previously the domain of nurses, pharmacists, and search engines. The regulatory response, if one comes, will likely start with what qualifies as medical advice under a chatbot's terms of service.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
