Apple opened enrollment on Monday for a research study that treats AI health advice as something to be measured rather than marketed. The Apple Health AI Usage Study, run with OpenEvidence, the clinical AI platform that generates answers from sources like the New England Journal of Medicine, wants up to 200,000 participants to ask real health questions in a new iOS app and then score what comes back on accuracy, detail and relevance. The app is downloadable now, arriving as Apple shutters another long-running way users got help.
Who can join and what they will do
The rules are narrow. Participants must be 21 or older and based in the United States, and the study requires an iPhone 15 Pro running iOS 26 or later, even though the app itself does not use Apple Intelligence. Questions go in as text; dictation through the iOS keyboard is available, but there is no Siri integration and no Apple Intelligence plumbing anywhere in the flow. Answers come back in plain language with links to the peer-reviewed clinical sources behind them, and the participant's job is to grade how well each response holds up.
During enrollment, users can opt in to sharing HealthKit data such as activity, sleep or clinical lab results, which lets the study see whether answers land differently when the system knows something about the person asking. Apple says the collected information stays confidential and will not be used to train AI models, a line it drew explicitly in the study materials, and one that separates this exercise from the data-hungry product pipelines users have learned to suspect.
Why Apple is measuring instead of shipping
The framing from Apple's health team is that nobody yet understands how people actually engage with, comprehend and evaluate health AI in daily life, and that a dedicated study is the honest way to find out. Sumbul Desai, Apple's vice president of Health and Fitness, put the company's position in a statement about responsible innovation starting with scientific rigor and clinical grounding, and about insights that should feed the broader scientific conversation rather than a product roadmap.
That caution tracks with how the company has handled health features on hardware, where claims have generally arrived after validation, as recent generations of the Apple Watch have shown with features that ship only when the evidence supports them. The device restrictions are worth reading the same way: the iPhone 15 Pro floor excludes a large share of existing users, which keeps the study population on hardware capable of running the app cleanly for the length of the research rather than maximizing enrollment convenience. The study's outputs are intended to improve the design of AI-powered health tools so they deliver information people can understand and act on, which is the question every general-purpose assistant now faces as users bring medical worries to chatbots never validated for the task. Whether that ends in an Apple health assistant is a question for later; for now the company is paying 200,000 volunteers to tell it where conversational AI in medicine breaks down.