Apple Launches Health AI Study App With OpenEvidence for 200,000 iPhone Users
On this page 7 sections
TL;DR
Apple shipped a study, not a product. The Apple Health AI Usage Study went live in the United States on Sunday, Nov 15: a free research app designed by Apple, built with clinical AI company OpenEvidence, and open to up to 200,000 people age 21 and over for more than 12 months. Participants ask everyday health questions and get answers written in plain language, grounded in The New England Journal of Medicine, JAMA and Nature, with linked peer-reviewed citations on every response. Sharing HealthKit data is optional, Apple says study data will not be used to train AI models, and OpenEvidence deletes the raw queries and responses after 24 hours.
What is actually on the App Store
The announcement landed on stage at the HLTH conference in Las Vegas and in Apple's newsroom the same day. The study app is free and enrolls US participants 21 and older who have an Apple Intelligence-enabled iPhone, meaning iPhone 15 Pro or later, running iOS 26 or later.
Inside, it works like a health chatbot with a bibliography. Participants ask about sleep, heart health, or anything everyday; the app answers in approachable language, and each response carries linked citations to peer-reviewed clinical literature. Users rate the answers and get monthly surveys. They can also opt in, per metric, to sharing HealthKit data such as resting heart rate, activity, sleep or lab results, so researchers can test whether personal context makes answers more relevant. A visit-prep summary rounds up what to raise at the next doctor's appointment.
"At Apple, we believe responsible innovation starts with scientific rigor and clinical grounding," said Sumbul Desai, M.D., Apple's vice president of Health and Fitness. OpenEvidence CEO Daniel Nadler pitched the pairing more economically at HLTH: "OpenEvidence is the most trusted and loved brand product among doctors. Apple is the most trusted and loved brand and product among consumers, and especially health consumers. So you have to bring these two things together. It's like peanut butter and jelly."
The privacy plumbing is the product
The design details are the most copyable part of the whole launch. Apple states flatly that study data will not be used to train AI models; it told TechCrunch the data is anonymized before it reaches a secure third-party cloud provider, and that OpenEvidence deletes the query and response data after 24 hours.
Consent is granular. Each HealthKit type is a separate decision, survey responses are separate again, and Apple says it has security measures around the study data it keeps. Guardrails sit inside the loop: the app is intended to detect certain concerning questions and suggest the participant follow up with a provider or other resources. The disclaimer is explicit: the app "is not meant to diagnose or treat any condition, or replace the advice of medical professionals."
Why a study, and why OpenEvidence
The market context explains the wrapper. OpenAI said in July that ChatGPT users were asking more than 300 million health-related questions a week, and both OpenAI and Anthropic have since shipped health-specific modes. Apple's answer to that demand is a consent form with a chatbot attached.
Shipping AI inside a study is the pop-up restaurant of health tech: real food, real diners, and nobody has to call it a restaurant yet. Apple gets to measure trust, comprehension and answer quality for a year, publish the findings, and only then decide what graduates into a product. Desai said as much on stage: "We're studying what builds user trust, how people comprehend responses, the quality of responses, and we wanted to work with a clinically grounded partner, and that is OpenEvidence."
The clinicians' side of the pipeline already exists at scale. A JAMA study published Oct 15 counted 952,008 US clinicians sending OpenEvidence 271 million queries between January 2024 and June 2026. UCSF's Robert M. Wachter, chair of the Department of Medicine, framed the Apple study as the other half of that story: "When patients and doctors draw on versions of the same trustworthy tools, patients arrive prepared, conversations improve, and next steps happen seamlessly."
What builders should copy
- Citations as interface. Every answer carries linked peer-reviewed sources, so a user can check the grounding instead of trusting the vibe. Any domain with a canonical literature can do this.
- A one-sentence privacy story. No training on study data, anonymized before a third-party cloud, deleted after 24 hours. If your data policy needs a paragraph of exceptions, it is not a story.
- Granular consent. Per-metric HealthKit opt-ins beat one binary toggle, and regulators notice the difference.
- Escalation before answers. Routing concerning questions to a provider is also how you stay on the safe side of the diagnose-and-treat line.
- Study as beachhead. In a regulated market, measure first and ship later. The study format buys a year of learning without pre-committing a product.
Read the small print
The study app is not a product launch. TechCrunch notes it is unclear whether any of this becomes a Health app or Siri feature, and Apple has promised neither. Participants self-select into the study, so the sample skews toward iPhone owners who opted in, and results are more than 12 months out. The short-term winner is OpenEvidence, now named on stage and in Apple's release as the study's clinically grounded engine.
Key Takeaways
- Apple launched the Apple Health AI Usage Study on Nov 15: a free US research app built with OpenEvidence, open to up to 200,000 participants age 21 and over, running for more than 12 months.
- Eligibility requires an Apple Intelligence-enabled iPhone (iPhone 15 Pro or later) running iOS 26 or later; enrollment happens through the App Store study app.
- Answers arrive in plain language with linked peer-reviewed citations from NEJM, JAMA and Nature; users rate responses, complete monthly surveys, and can share HealthKit metrics (resting heart rate, activity, sleep, lab results) per type.
- Privacy design: study data never trains AI models, data is anonymized before a third-party cloud provider, OpenEvidence deletes queries and responses after 24 hours, and guardrails route concerning questions to providers.
- Context: a JAMA study counted 952,008 US clinicians and 271 million OpenEvidence queries through June 2026; Apple is now pointing the same clinical engine at consumers, inside a study.
- Caveats: this is research, not a shipped product, with no commitment to a Health or Siri feature; the sample self-selects, and results are more than a year away.
Sources: Apple Newsroom, Nov 15, 2026, Apple Health AI Study App, App Store, STAT, Nov 15, 2026, TechCrunch, Nov 15, 2026, TechCrunch, July 23, 2026, JAMA: Clinician Use of an AI-Enabled Clinical Information Platform