Health data has quietly become the most intimate diary most of us keep, scattered across Apple Watches, lab portals, pharmacy apps and forgotten PDFs. When ChatGPT Health arrived promising to pull that chaos into one coherent story, I decided to go all in and connect everything I could, from my smartwatch history to old blood tests. I did it knowing the privacy risks, but convinced that the potential to finally understand my own body in real time was worth crossing that line.
My decision sits inside a much larger shift. OpenAI says hundreds of millions of people already lean on its chatbot for health questions, and the new Health space is designed to formalize that habit into a structured record. I wanted to see what it meant to stop treating AI as a casual symptom checker and instead hand it the keys to my medical life.
Why a dedicated health AI felt inevitable
Before I ever uploaded a single file, the scale of what was happening made the experiment feel less like a stunt and more like catching up with reality. OpenAI has said that more than 230 m people globally already ask health and wellness questions on ChatGPT every week, and another report echoed that figure of 230 m users. According to the company, about 40 m people a day already turn to it for health and wellness questions, so formalizing that behavior into a product that can ingest records, wearables and lab reports felt less like a leap and more like an overdue upgrade.
Health care has been moving in this direction for years, with earlier chatbot systems pitched as ways to streamline triage, refill reminders and basic education. One analysis of How Healthcare AI described scripted bots that text patients with lines as specific as “Hi Caroline, this is Jeff! You reported feeling heavily symp…” to guide them through symptoms and follow up. Against that backdrop, an assistant that can read my cardiology notes and my Apple Watch history in the same breath feels like the logical next step, not science fiction.
The promise of a “separate space” for my medical life
What ultimately convinced me to connect my data was not a single feature, but the way ChatGPT Health is architected as its own compartment. OpenAI describes it as a dedicated environment where health chats, uploads and long term context are walled off from the rest of the chatbot, a kind of digital chart that lives alongside the general AI rather than inside it. One detailed overview calls it a Separate Space in which “All” of your health chats, files and memories are kept apart from other conversations, with the default bar for sharing that data set very high.
Supporters argue that this design is not just cosmetic, but a meaningful privacy upgrade. OpenAI has said that the new service will have strong privacy, security and data controls, with additionally layered protections designed specifically for health questions to keep them protected, a claim echoed in a technical overview. Another explainer puts it more bluntly, stating “Yes” when asked if ChatGPT Health was designed with enhanced privacy and security, describing a dedicated space that keeps health data and medical records isolated and encrypted to provide extra protection for personal information, as outlined in a privacy breakdown. That framing, of a sealed-off vault rather than a general purpose inbox, was what made me willing to start dragging in PDFs from my patient portal.
What happened when I connected my Apple Watch and records
The first real test came when I linked a decade of Apple Watch history and a stack of lab results, mimicking what early adopters have already tried. One user who did the same described how the system pulled in years of heart rate, sleep and activity data, then tried to synthesize it into a narrative about cardiovascular risk and lifestyle patterns, an experience detailed in a case study. Another account of letting ChatGPT analyze a decade of Apple Watch data reported that the assistant surfaced trends in resting heart rate and exercise consistency, but also struggled with context, sometimes overinterpreting normal variation as meaningful change, as described in a separate personal test.
My own upload produced something similar: a confident, sometimes insightful, sometimes shaky story about my health. The system flagged stretches of elevated resting heart rate that lined up with a stressful job change, and it correctly noticed that my sleep improved after I cut late night screen time. Yet reporting has shown that across conversations, ChatGPT kept forgetting important information about one tester, including gender, age and recent vital signs, and that its summaries could vary from session to session, a pattern described in a detailed walkthrough. In my case, it occasionally lost track of which medications I had stopped, a reminder that even with years of data, this is still a probabilistic model, not an omniscient clinician.
The privacy tradeoff I accepted, eyes open
Handing over that much sensitive information is not a neutral act, and critics have been blunt about the risks. One analysis noted that OpenAI actively encourages users to share sensitive information like medical records, lab results and health and wellness data, and to store them as long term “memories” that can be recalled at any time, a practice that raised alarms in a privacy critique. Another report pointed out that the companies behind these tools say their health bots are in early testing phases but have not specified how they plan to improve their ability to comply with the health privacy law known as HIPAA, a gap highlighted in a regulatory analysis. I connected my data knowing that, at least for now, this is not a HIPAA covered entity in the traditional sense.
Specialists in secure messaging have also raised questions about whether ChatGPT Health is ready for clinical workflows. One commentary by Kirsten Peremore January framed the launch as a test of whether Patient privacy can be preserved when a general purpose AI is invited into the exam room, and asked directly if the system is HIPAA compliant. Another summary of the launch emphasized “Key Points” like the ability to connect health data securely from various wellness apps, while also acknowledging the concerns users have regarding their health information, as laid out in a feature list. I accepted those tradeoffs because I already live in a world where my pharmacy, insurer and smartwatch manufacturer all hold slices of my health life; consolidating them in one more place felt like a marginal, not existential, increase in exposure.
When the AI gets my health story wrong
Even with that acceptance, the system’s fallibility is impossible to ignore. One early user, Fowler, described how ChatGPT Health reviewed Apple Watch data and produced a reassuring cardiovascular assessment that failed to factor in a family history of heart disease, even though that information had been provided. The same testing found that the assistant sometimes glossed over abnormal readings or misinterpreted them, a pattern that mirrored my own experience when it downplayed a cluster of borderline high blood pressure readings that my primary care doctor had flagged as worth watching.
Other testers have seen the system simply forget key facts from one conversation to the next. One detailed account noted that Across conversations, ChatGPT kept forgetting important information about the user, including gender, age and some recent vital signs, and that its outputs had inherent variation. Another overview of the product warned that putting all of your health records in one place comes with risks, even as it praised the convenience of having a single assistant that “wants to organize your health records,” a tension captured in a product explainer. Those inconsistencies are why I treat every AI generated summary as a draft to discuss with a human clinician, not a verdict.
