Apple Health AI Study: What Joining Actually Shares

Face-down phone beside a closed notebook and measuring tape on a wooden desk

The Apple Health AI study is a U.S. research app, not a doctor and not a general chatbot. Apple announced the study on November 15, 2026, to answer everyday health questions from cited medical literature, with Health data only if you opt in. Joining can share questions, surveys, and any HealthKit types you allow, and Apple says study data will not train AI models.

Updated November 16, 2026

Key Takeaways

  • Up to 200,000 U.S. adults 21+ can join on iPhone 15 Pro or later.
  • Cited answers do not diagnose, treat, or replace a clinician.
  • HealthKit sharing is optional; study data will not train models.
  • OpenEvidence deletes questions and responses within 24 hours.

What can the Apple Health AI study do, and not do?

Download the study app, complete informed consent, and ask text questions. The Verge reported that keyboard dictation works, but the app is not integrated with Siri or Apple Intelligence.

OpenEvidence, used by clinicians, returns answers in language those professionals are used to, drawn from sources such as The New England Journal of Medicine, JAMA, and Nature. ABC News reported that Apple’s own technology turns that research into easy-to-understand language with linked citations. CNET also listed the National Comprehensive Cancer Network and the Cochrane Database of Systematic Reviews.

Optional HealthKit data lets researchers test whether your context makes answers more relevant. The Verge said you rate responses for accuracy, detail, and relevance. MacRumors and ABC News reported that participants also complete surveys.

Apple says the app is not meant to diagnose, treat, or replace a clinician, and you should verify concerns with a doctor. There is no live monitoring. Guardrails are intended to detect certain concerning questions and suggest a provider or other resources. CNET reported staff will not contact you about self-harm questions.

What does joining actually share?

CNET reported you are asked for race, age, and education level, used to explore how those factors may affect preferences and communication needs. The iPhone strips identifying details; UST Global then delinks questions from participant IDs and screens them before OpenEvidence answers.

  • HealthKit types you choose, such as resting heart rate, activity, sleep, or lab results
  • Baseline, monthly, and doctor-visit surveys
  • Coded survey responses separated from personally identifiable information, plus conversation summaries and question categories, for a small Apple research group

Clinical records need a second share from the Health app. OpenEvidence deletes questions and responses within 24 hours and will not train on them, Apple told ABC News. CNET said the data may inform Apple features. Apple told ABC News it plans to make the research available broadly to help computer-assisted health communication. No publication date is set.

How does this compare with ChatGPT-style health questions?

A February 2026 Oxford study in Nature Medicine, cited by ABC News, found chatbots identified conditions nearly 95% of the time on clinical cases, yet everyday users reached the right answer less than 35% of the time. CNET said one in three U.S. adults use chatbots for health information. This analysis uses ChatGPT-style for that habit, not a product these sources review.

Question Study app General health chatbot
Grounding Linked citations from NEJM, JAMA, and Nature ABC: journals behind OpenEvidence may be unavailable to other AI platforms
Your data Optional HealthKit; records need a separate share Not part of this study
Training Not used to train models; questions and responses deleted within 24 hours Equivalent terms are not described in these sources

Citations and easier language are the study’s answer to that gap. These sources do not give a general chatbot the same enrollment of up to 12 months, 24-hour deletion, or non-training terms.

Is the cited-source model worth the HealthKit opt-in?

HealthKit is optional. The reason to opt in is answers that use your measurements, as in ABC’s resting-heart-rate example. In our reading, citations do not require that share. Clinical records take a separate yes. Unlike a Mac knowledge graph that has no off switch, this share can stay off.

Up to 200,000 U.S. adults age 21 or older can join with an iPhone 15 Pro or later on iOS 26 or later, for up to 12 months. A paired Apple Watch can prompt check-ins, and a complete lack of participation can lead to removal, CNET reported. The Verge said the app does not use Apple Intelligence and is not integrated with Siri; CNET, citing Apple, said on-device models classify question categories.

Our view: treat the app as visit prep. Robert M. Wachter, M.D., of UCSF, told Apple that patients arrive prepared when they and their doctors draw on versions of the same trustworthy tools. Leave HealthKit off unless personal context is why you joined. There is no live monitoring, and the app does not replace a clinician.

Frequently asked questions

Is the Apple Health AI study a doctor?

No. Apple says the app is not meant to diagnose or treat any condition or replace medical professionals, and there is no live monitoring.

Will joining train an AI model on my questions?

Apple says study data will not be used to train AI models. OpenEvidence deletes question-and-response data within 24 hours and will not train on those queries.

Do I have to share Apple Health data?

No. HealthKit metrics are optional, and clinical records require a separate share from the Health app into the study app.

Who can enroll in the study?

U.S. adults age 21 or older with an iPhone 15 Pro or later on iOS 26 or later, up to 200,000 people. CNET said the ask runs up to 12 months.

Notice an error?

Help us improve our content by reporting any issues you find.