Apple Launches Apple Health AI Study to Test How Users Trust and Act on Health Chatbot Answers

No live monitoring, and no outreach if someone asks about self-harm
Apple's study app has safety guardrails, but staff do not watch conversations or contact participants.
Mark

Why does Apple need a study for this? People already ask chatbots health questions.

Mimi

That's the point. One in three US adults already does, so Apple wants to know whether people understand the answers, trust them, and act on them.

Mark

And the answers come from OpenEvidence, not a general chatbot?

Mimi

Yes, a platform built for doctors, drawing on sources like NEJM, JAMA and Cochrane. Answers carry linked peer-reviewed citations.

Luke

Worth noting that's a description of where the information comes from, not proof the answers are accurate. The study measures trust and comprehension, not clinical correctness.

Mark

What about privacy? That's a lot of sensitive data.

Mimi

Questions are scrubbed on the phone, delinked at UST Global, then sent to OpenEvidence, which deletes data within 24 hours. HealthKit sharing is optional and takes two separate choices.

Luke

Those are Apple's stated safeguards. Nothing here independently verifies them.

Mark

And the safety side?

Mimi

Guardrails suggest following up with a provider, but there's no live monitoring, and staff won't reach out if someone asks about self-harm.

Luke

That's a notable limit. And Apple says data won't train AI models but may inform feature development. How far that goes is unknown.

  • Apple has released a dedicated app that asks 200,000 American participants to put their health questions to an AI and then report honestly on the answers.
  • Roughly one in three US adults already turns to chatbots for health information, so the study arrives into a practice that is spreading faster than anyone has measured it.
  • Questions pass through layers of stripping, relabelling and screening on phone, cloud and third-party servers before reaching the OpenEvidence system, with all question data deleted within 24 hours.
  • Participants who choose to share sleep, lab and activity data receive more personalized answers, while the study tracks whether comprehension, quality and trust actually hold up over twelve months.
  • The findings will not train AI models but may steer Apple's own feature decisions, and the industry is watching to see whether this becomes a template for responsible health AI.

On a quiet Sunday, Apple opened a door that many people have already walked through on their own: asking an artificial intelligence about their bodies, their symptoms and their test results. The Apple Health AI Study is a year-long inquiry into that habit, drawing on a decade of the company's research partnerships with medical institutions, and it asks a question that is older than any software: when someone seeks answers about health, what makes them believe what they are told? Its answers will shape how the next generation of health tools is built, and in that sense the study is less a product launch than an attempt to learn, in public and with care, what it means to be helped.

On Sunday, Apple released an app with a single purpose: to let people ask a health chatbot questions and then ask them how it went. The Apple Health AI Study is the company's newest research project, and it lives in its own dedicated app rather than inside an existing one. It continues a line of work that began in 2019 with Apple's Research app and its partnerships with medical institutions, which previously covered hearing, heart health, movement and women's health.

The study asks four things: whether people understand what the AI tells them, how good the answers are, what builds or erodes trust, and whether the advice leads to everyday health actions. The motivation is simple. Many people already use AI chatbots for health information, and Apple wants to learn how to design such tools responsibly, for itself and the wider industry. Answers come from OpenEvidence, a platform built for doctors that draws on journals such as the New England Journal of Medicine and JAMA, with linked citations for every response.

Participants must be US adults aged 21 or older with an iPhone 15 Pro or later. They can stay in the study for up to a year, rating each answer and describing what worked. They fill out surveys at the start, monthly, and around doctor's visits, and they can optionally share HealthKit data such as sleep, lab results and heart rate for more personal responses. Onboarding requires informed consent and demographic details, which Apple says will help it understand how background shapes what people need.

Privacy runs through the design. Questions are stripped of identifying details on the phone, passed through a third-party cloud environment, and then delinked from participant IDs before reaching OpenEvidence, which deletes data within a day. Apple researchers see only coded responses and summaries. The app flags concerning questions and points people toward providers or other help, though nobody monitors conversations live, and staff will not follow up with people who ask about self-harm.

Apple says the data will not train its AI models, but it may shape future features. Whether the study becomes a template for the industry depends on what it finds about how people actually think when a machine answers their most personal questions.

On Sunday, Apple released a new app that does one thing: it lets people ask a health chatbot questions and then asks them how it went. The Apple Health AI Study is the company's latest research project, and it arrives with its own dedicated app rather than living inside an existing one.

The project builds on a line of work that began in 2019, when Apple introduced its Research app and started partnering with medical universities and institutions. Earlier studies covered hearing, heart health and movement, and women's health. The most recent before this one, the Apple Health Study, spans a broad range of health areas. According to what Apple said in a press briefing, the new study is aimed at how people interact with conversational health AI tools, and it will look at four things: comprehension, response quality, the drivers of user trust, and everyday health actions.

Apple frames the purpose as informing the responsible design of health AI, both for itself and for the wider industry, and exploring whether such tools can help people prepare for doctor's visits. The questions at the center are plain: do users understand what the AI tells them, do they trust it, and is it useful. The timing reflects a habit already in place. One in three US adults uses AI chatbots for health information. Apple says study data will not be used to train AI models, though it may inform the company's feature development.

To supply the answers, Apple is working with OpenEvidence, an AI platform built for doctors. It draws on medical information from The New England Journal of Medicine, the Journal of the American Medical Association, the National Comprehensive Cancer Network, Nature and the Cochrane Database of Systematic Reviews. Inside the app, participants type health questions and receive answers with linked, peer-reviewed clinical citations. They can optionally share HealthKit data, including sleep, lab results, activity and resting heart rate, to get more personalized responses. HealthKit is the hub for health and fitness data gathered on the iPhone and Apple Watch.

The study is open to 200,000 participants, who must be in the US, at least 21 years old, and using an Apple Intelligence-enabled iPhone, meaning an iPhone 15 Pro or later, running iOS 26 or later. Apple says the app needs those phones' processing power and uses on-device Apple Foundation Models, which in the study classify question categories.

Participation can last up to 12 months. People can ask about general health, specific symptoms, recovery from a procedure, or their own uploaded data. After each answer they rate it and describe what worked and what could be better. There is a baseline survey, monthly surveys, and extra surveys tied to doctor's visits, with follow-ups afterward on how the visit went and whether the preparation tools helped. Apple Watch owners get check-in notifications with suggested topics. A participant who stops engaging entirely may be removed.

Onboarding is mandatory and includes informed consent, plus demographic details such as race, age and education level, which Apple says will be used to explore how those factors shape preferences and communication needs. Questions are first scrubbed on the iPhone of identifying details, then passed to a secure cloud environment run by a third-party contractor, UST Global, where they are delinked from participant IDs and screened again. Only then do they reach OpenEvidence, which generates answers and deletes both question and response data within 24 hours. A small group of Apple researchers can see coded survey responses without personal identifiers, conversation summaries, and categories of common conversation types.

The app has guardrails that detect concerning questions and suggest following up with a provider or other resources. There is no live monitoring, and study staff will not contact participants who ask about self-harm. Sharing health records is a two-step choice: first uploading clinical records to Apple's Health app, then separately sharing them with the study app. What Apple learns from the findings may shape both its own features and the broader industry's approach to health AI.

The study aims to understand how people interact with conversational health AI tools, examining comprehension, response quality, trust and everyday health actions.
— Apple, during a press briefing
Contact Us FAQ