OpenAI has presented ChatGPT Health. This is a separate service that combines personal health information and generative AI within a strictly defined and secure environment. The new feature is intended to better inform and prepare users for conversations about health and well-being, without replacing the role of healthcare professionals.
Health is already one of the most discussed topics within ChatGPT. In recent months, there has also been much debate about the benefits and risks of using generative AI in healthcare, particularly in mental health care. Concerns have been raised about the “approach” (answers) to questions from people struggling with mental health issues, for example. Other studies, however, have emphasised the benefits of AI in mental health care. The fact is that hundreds of millions of people worldwide ask questions about health and lifestyle every week. With ChatGPT Health, OpenAI says, this is being built upon, with extra attention to privacy, security and context. The feature makes it possible to base conversations on personal health data, such as medical records and data from wellness apps, making answers more relevant and practical.
Developed with healthcare professionals
ChatGPT Health was developed in close collaboration with doctors and healthcare professionals. The functionality is explicitly supportive: intended to provide insight, recognise patterns and enable users to be better prepared for conversations with healthcare providers.
In current practice, health information is often fragmented across patient portals, apps, wearables, PDFs and separate notes. This makes it difficult for patients to maintain an overview and see the bigger picture. ChatGPT Health therefore introduces a separate, dedicated environment within ChatGPT in which health-related conversations, files and linked data sources remain strictly separate from other chats.
Users can, of course only with explicit consent, link electronic patient records and wellness apps, including Apple Health, Function and MyFitnessPal. This allows ChatGPT to assist in interpreting test results, preparing consultations, analysing lifestyle data and making decisions about nutrition, training or health insurance.
ChatGPT Health is not intended for diagnosis or treatment. The focus is on everyday questions, long-term insights, and support outside of acute care moments.
Privacy and security
Health data is among the most sensitive personal data. That is why ChatGPT Health functions as a separate space with additional security layers on top of ChatGPT's existing privacy and data controls. Conversations and files are stored and transmitted in encrypted form by default. Additional isolation and specialised encryption have been applied specifically for health data.
Importantly, conversations within ChatGPT Health are not used to train AI models. Health information remains within this defined environment and is not shared with other chats. Users retain full control over their health memory and can view or delete it at any time via the settings. For additional security, multi-factor authentication (MFA) can be enabled.
Link to external data sources
To unlock electronic medical records, OpenAI is collaborating with b.well, an American network for securely connecting health data. These links are currently limited to the US and available to users aged 18 and older. Apple Health integration requires iOS.
Wellness apps can also only be linked with explicit consent. Apps that access ChatGPT Health must meet additional privacy and security requirements, including data minimisation and an additional security review. Users can terminate links at any time, after which access is immediately revoked.
The development of ChatGPT Health has been supervised by more than 260 doctors from 60 countries and a wide range of specialisms. In total, they provided more than 600,000 pieces of feedback on AI responses, covering 30 medical areas of focus. This input was used to train the model in clear communication, risk awareness and the timely encouragement of contact with healthcare professionals.
The quality of responses is assessed using HealthBench, an evaluation framework developed in collaboration with practising doctors. Instead of theoretical knowledge questions, the emphasis is on safety, contextual understanding, comprehensibility and clinical applicability.
Phased introduction
ChatGPT Health will initially be made available to a limited group of users. This phased rollout will allow feedback to be collected and the functionality to be further refined. In the coming weeks, access will be extended to more users on the web and iOS. Interested parties can sign up for a waiting list. However, this is not yet possible in the Netherlands at this time.
