OpenAI has handed out a key to all users in the U.S. who are 18 years old and above today — officially removing the waitlist and fully launching the new Health (Health) feature. Whether you use the free version, Go, Plus, or Pro version, you can directly turn on this experience in the ChatGPT sidebar on both the web and iOS platforms.

This demand has been rising for a while. According to OpenAI's statistics, the number of users asking health and wellness-related questions on ChatGPT has risen from 230 million in the early stages to over 300 million. Based on feedback from several months of testing, this health experience has undergone multiple upgrades: the system can now connect with Apple Health and read electronic medical records from One Medical, Function Health, and several hospital systems in the U.S.

image.png

After user authorization, ChatGPT can combine data such as medication, lab results, medical records, sleep, and daily exercise to analyze and provide truly personalized health answers — for example, comparing your latest lab results with previous data, summarizing physical changes since your last visit, or assessing the impact of sleep and exercise on your daily condition.

Interestingly, during early testing, OpenAI found that many people did not always ask questions in the dedicated health section. Therefore, ChatGPT now supports using health data outside the dedicated area: users can either type "@Health" to explicitly bring in the health context, or authorize ChatGPT to refer to personal health background when answering everyday questions — automatically avoiding dietary restrictions when recommending meals, or avoiding recent injury areas when planning physical activities. By default, the system will always ask for permission before accessing linked health data, and users can choose single-use authorization or permanent permission.

On the issue of privacy, OpenAI made it clear in its official statement: all medical records, Apple Health data, and conversation content generated using these data will not be used to train large language models or for targeted advertising. Bringing your medical records to a chatbot to read them has raised both the threshold and the trust factor, which have now been put on the table.