A smartphone resting on a wooden desk showing a glowing abstract chat bubble, with a stethoscope lying beside it in warm morning light
News

Apple Will Study How 200,000 Americans Talk to a Health Chatbot for a Year

Apple and OpenEvidence launched a study app for up to 200,000 Americans to use a clinically grounded health AI for 12+ months, rating answers for trust and comprehension — a slow, research-first answer to the 300-million-a-week health query flood.

AppleOpenEvidencehealth AIchatbotsmedical advice

Apple announced on Sunday the launch of the Apple Health AI Usage Study — a research programme that ships no product and makes no promises. It will simply watch, at unprecedented scale, what happens when ordinary people ask a chatbot their health questions. Up to 200,000 US participants aged 21 and over, each with an iPhone 15 Pro or later running iOS 26, can enrol in a study app built with clinical-AI firm OpenEvidence, as Apple’s newsroom states. The study runs more than 12 months.

🔍 THE BOTTOM LINE

The most valuable number in consumer AI right now is one nobody publishes: how often people trust a chatbot’s answer about their own body. Apple’s study is the first attempt to measure that relationship at population scale, with the explicit framing that LLM health responses “may at times” be incomplete or unclear — and while OpenAI counts more than 300 million health-related queries per week crossing its servers, as TechCrunch notes, no vendor has actually studied usage with clinical rigour before. Apple is choosing to be the measured exception in a market that is otherwise racing to answer first.

What participants actually do

The app works like ChatGPT with a citation layer. Participants ask everyday health questions — the same questions they already ask general chatbots — and receive answers grounded in peer-reviewed medical literature, with each response linked to sources such as the New England Journal of Medicine, JAMA and Nature, per Apple’s announcement. They then rate the responses on criteria including accuracy, detail and relevance, as The Verge describes, and complete monthly surveys.

Optional layers go deeper. Participants can share HealthKit data — resting heart rate, activity, sleep, lab results — to see whether personalised context improves the answers’ relevance, and the app is designed to help people prepare for doctor’s visits rather than replace them: safety guardrails are intended to detect concerning questions and steer participants toward a provider.

The trust-and-comprehension mission

Apple’s stated purpose is not a product launch. The study examines what drives user trust, the quality of responses, comprehension, and everyday health actions, with findings meant to inform design for Apple and the industry, per the newsroom release. Apple VP of Health Sumbul Desai framed it as rigour first: “Health is deeply personal, and conducting a dedicated research study is the most thoughtful way to understand how people engage with, comprehend, and evaluate health AI in everyday life.”

The privacy design is the part that will get scrutinised. Apple says study data will not be used to train AI models, participants control whether to share HealthKit metrics or survey responses, and — per TechCrunch — OpenEvidence deletes query and response data after 24 hours. The study is also a not-subtle positioning move: Apple’s app requires an Apple Intelligence-capable iPhone but does not itself use Apple Intelligence, is text-only, and is not integrated with Siri. Apple is studying the space before committing to it.

Why OpenEvidence is the interesting part

The partner, not the platform, is the story here. OpenEvidence is the clinical AI platform already widely used by healthcare professionals, which Apple describes as generating answers from established sources including the New England Journal of Medicine, JAMA and Nature, per Apple’s release. UCSF’s Robert Wachter, quoted in the release, sketched the vision: “Patients and clinicians are both turning to AI, but the tools they use are being built and tested in separate worlds… When patients and doctors draw on versions of the same trustworthy tools, patients arrive prepared, conversations improve, and next steps happen seamlessly.”

That is a different thesis from the consumer-lab race. OpenAI, Google and Anthropic are bolting health features onto general chatbots for hundreds of millions of users. Apple is testing whether the clinically grounded, citation-forcing tier — the one physicians already use — can translate to patients. If it can, the trust gap that general chatbots carry becomes an obvious market; if people find cited answers slower and less reassuring than confident ones, the consumer-lab approach wins by default, and the study will have documented why.

The New Zealand angle

The study is US-only, but the pattern it measures will reach here faster than the app. New Zealand’s health system already has a documented AI-advice problem: Health NZ clinicians were found using shadow AI for clinical notes, as we covered in our Health NZ shadow-AI piece, and health misinformation has been a live issue since the TikTok fake-doctor wave we covered in AI fake doctors. New Zealanders’ health questions already flow to general chatbots, and there is no local study of usage patterns at anything near this scale, so whatever Apple measures at 200,000-person scale becomes useful evidence for regulators everywhere, including here.

❓ FAQ

What is the Apple Health AI Usage Study? A virtual research study launched Sunday, 15 November 2026, run by Apple with clinical-AI firm OpenEvidence. Up to 200,000 US participants aged 21+ ask health questions in a study app and rate the AI’s answers on accuracy, detail and relevance, plus monthly surveys, over more than 12 months.

Do I need Apple Intelligence for the study app? You need an Apple Intelligence-enabled iPhone 15 Pro or later running iOS 26 to enrol, but per The Verge the app itself does not use Apple Intelligence, is text-only, and is not integrated with Siri.

Is my health data used to train AI? No — Apple states study data will not be used to train AI models, participants choose whether to share HealthKit data, and OpenEvidence deletes query and response data after 24 hours, per TechCrunch.

Why does this matter for health AI? It is the first population-scale study of how ordinary people actually use, trust and act on conversational health AI, with answers grounded in peer-reviewed sources. That evidence gap sits behind every regulator’s question about chatbot medical advice.

When does it reach New Zealand? The study is US-only. But the usage and trust patterns it documents are exactly the evidence MedSafe and NZ health agencies lack as New Zealanders already use general chatbots for health questions.

🔍 THE BOTTOM LINE

The tell in Apple’s announcement is what it declines to do: no product integration, no Siri, no feature ship. The most valuable consumer-AI dataset of 2026 — how people respond when a chatbot answers a question about their own body — is being collected by a company that can afford to wait a year before deciding what to build. Trust, not capability, is the scarce resource in health AI, and Apple just started measuring the supply.

📰 Sources

  • Apple Newsroom — Apple launches research study to explore the use of AI for health (15 November 2026)
  • TechCrunch — Apple launches new app to study how people use AI for health (15 November 2026)
  • The Verge — Apple’s new study looks to improve the effectiveness of AI health tools (15 November 2026)