Apple Launches AI Health Study With OpenEvidence to Track How Users Ask Medical Questions

Apple has launched a research study app, built with healthcare AI company OpenEvidence, to examine how people use AI for health questions. The study is open to U.S. users over age 21 with an iPhone 15 or later running iOS 26 or later and Apple Intelligence enabled. TechCrunch reported the launch on November 15, 2026, and the app became available to download for U.S. users that day.
Apple's aim is to run the study with up to 200,000 participants over more than 12 months. The app gives simplified answers to everyday medical questions. Those answers are grounded in clinical data from sources including The New England Journal of Medicine, The Journal of the American Medical Association, and Nature. Beyond answering questions, the app can help participants prepare for a doctor's visit and direct them to specific providers when a medical issue looks concerning.
Participants can rate answers and leave feedback on individual responses, and they are asked to complete monthly surveys. For more personalized suggestions, they can share HealthKit data, the health metrics stored on the iPhone, including resting heart rate, activity, sleep, or lab results.
Apple said study data will be anonymized, meaning identifying details are stripped out, before it is sent to a secure third-party cloud provider. Apple also said the data will not be used to train AI models. It added that OpenEvidence deletes query and response data after 24 hours. Apple's Newsroom lists "Apple launches research study to explore the use of AI for health" as its most recent item.
What sets the study apart from a standard consumer health feature is its design. The app is a deployed product and also a measurement instrument. Each query, rating, and monthly survey response becomes a data point on how users phrase health questions, how they weigh a clinical-grade answer against their own symptoms, and whether a provider referral changes what they do next. Over a planned run of more than a year, that record could show more about the gap between an answer and an action than a single benchmark, a standardized test run on a curated question set.
The broader context here is the data path. Health questions rank among the most sensitive inputs a consumer can give a model, and the study pairs them with HealthKit signals that are more granular than a typed symptom. Apple's published terms rest on two commitments: anonymization before transfer to a third-party cloud, and no use for model training. The 24-hour deletion window applies to OpenEvidence's handling of queries and responses. A technically literate participant will want to test those claims against the eventual protocol and privacy documentation, because the study's credibility depends on them.
In my view, the most useful question is not whether the answers are accurate in isolation, but whether they change behavior in a measurable way. A tool that nudges someone to see a doctor sooner, or helps them arrive at an appointment with better questions, offers something different from one that simply answers. The monthly surveys and feedback mechanisms are designed to capture that difference. Whether they do so with enough rigor will become clear only in the results Apple eventually publishes.
For now, the direct effect reaches only those who enroll. Eligible users can join today, and Apple's stated horizon is more than a year of data collection. The larger possibility is that a consumer AI product can be studied in the open, with defined data handling, clinical sourcing, and a stated research purpose. If the study delivers on that design, it could give the health AI field a concrete record of how ordinary people use these tools, a dataset the field has largely had to infer until now.


