All posts

Apple launched an AI health chatbot as a study. Here is what it is really testing.

Apple launched the Apple Health AI Usage Study with OpenEvidence. Who can join, what the app does, and why Apple is measuring what people do with the answers.

Apple launched an AI health chatbot as a study. Here is what it is really testing.
On Sunday, November 15, Apple announced the Apple Health AI Usage Study. It is a research study, run through a new iPhone app, where people ask an AI health questions and get answers with the medical research linked underneath.
Most of the coverage calls it Apple's AI health chatbot, and that is a fair description. I build a health coaching app, so I read announcements like this closely. What stayed with me was the list of things Apple decided to measure.

What Apple launched

The study is open to up to 200,000 adults in the United States who are 21 or older. You need an iPhone 15 Pro or later running iOS 26. You download the Apple Health AI Study App, go through a consent process, and start asking questions. The details below come from Apple's announcement and ABC News.
  • The answers are built on OpenEvidence, a clinical AI tool that Apple describes as widely used by healthcare professionals. It draws on sources like The New England Journal of Medicine, JAMA and Nature.
  • Each answer comes with citations to the research behind it.
  • You can choose to share data from the Health app, such as resting heart rate, activity, sleep or lab results, so the answer is about your own numbers.
  • You rate the answers and fill out a survey now and then.
ABC gives the example of someone asking how to improve their resting heart rate and getting advice based on their own measurements instead of a general recommendation.
Apple is clear about the limits. Its release says the app "is not meant to diagnose or treat any condition, or replace the advice of medical professionals."
On privacy, Apple says study data will not be used to train AI models. ABC reports that identifying information is removed from a question before it reaches OpenEvidence, and that OpenEvidence deletes the question and the answer within 24 hours.

Apple is studying what people do with the answer

One sentence in Apple's announcement matters more to me than the rest. The study will look at "what drives user trust, the quality of responses, comprehension, and everyday health actions."
Only one of those four is about whether the AI is right. The other three are about the person. Do they believe the answer? Do they understand it? Do they do anything differently afterward?
Apple has a good reason to ask. In February, researchers at Oxford published a randomized trial in Nature Medicine with 1,298 people in the UK. Each person got a medical scenario and had to work out what might be wrong and what to do about it. Some of them used a chatbot (GPT-4o, Llama 3 or Command R+). The rest used whatever they normally would.
  • Tested on their own, the chatbots identified the right conditions in 94.9% of cases.
  • People using those same chatbots identified a relevant condition in fewer than 34.5% of cases.
  • The chatbot group did no better than the control group.
The models knew the medicine. The knowledge got lost somewhere in the conversation. People left out details the model needed, and the answers changed with small differences in how a question was worded.
Dr. John Brownstein of Harvard Medical School said it plainly to ABC News: "Millions of people are turning to AI for health advice, but accurate answers aren't always useful."
And a lot of people are asking. Pew Research Center surveyed 3,488 US adults in June and found that 34% have used an AI chatbot for at least one health reason. A quarter have used one to figure out what is causing their symptoms.
So I read this as Apple being careful, and I respect it. Apple could have shipped a health chatbot to every iPhone. It chose to find out first whether the answers help anyone.

The doctor's tool is now in the patient's pocket

OpenEvidence was built for clinicians, and ABC notes that full access is still limited to healthcare professionals. In this study, answers from that same evidence base land on the patient's side of the exam table, with the patient's own sleep and lab results attached.
I wrote last week about Oura agreeing to buy Counsel Health. Apple's study is the same story from a different company. Wellness and healthcare used to be separate. The ring and the phone sat on one side, and the doctor and the lab order sat on the other. Now the consumer products are reaching across. Oura is buying physicians. Apple is borrowing the physician's reference tool.
Apple also says where it wants this to go. The study will explore how health AI can help people "better prepare for clinical visits with their own doctors." Dr. Sumbul Desai, Apple's vice president of Health and Fitness, told ABC that studying these experiences is essential "to ensuring AI serves as a bridge, not a barrier, to better care."
I agree with that direction. I have argued before that the better role for AI in healthcare is to make a professional's care easier to understand and follow. A patient who walks in with clear questions gets more out of a short visit.

What a correct answer cannot do

Go back to the resting heart rate example. Suppose the app reads your data and gives you a careful, cited answer about your own numbers. It will probably say some version of this: move more, sleep more, drink less.
You knew most of that already. I wrote about the same thing when someone fed eight years of Fitbit data into an AI and learned that staying up late costs them sleep.
Resting heart rate comes down over months of doing those things on ordinary days, including the days you do not feel like it. The answer cannot do that part for you, however good the citations are.
I learned this the expensive way. I paid an online coach $400 a month, and he texted me about my health every single day. I already knew the basics. I was paying for his attention.
That is why "everyday health actions" is the phrase I would watch in Apple's results. It is the hardest of the four to move. A question-and-answer app waits for you to open it, and the days that decide your health are usually the days you don't.
A pair of walking shoes on a doormat by the front door in morning light
A pair of walking shoes on a doormat by the front door in morning light

What I would watch for

I have a few reservations, and they are worth saying out loud.
  • The people who enroll will not be typical. They own a recent iPhone Pro, they volunteered for a health study, and they are comfortable asking an AI about their body. Whatever Apple learns about trust and comprehension will come from that group first.
  • A study is not a product. Apple told ABC the goal is to make the research broadly available, as well as to inform its own future products. Nothing in the announcement says a health chatbot is coming to every iPhone, or when.
  • Trust cuts both ways. Brownstein told ABC the open question is whether people "trust them appropriately." An answer with journal citations under it looks authoritative. That helps when it is right and you understood it. It hurts when you take it as a reason to skip the appointment.
I don't hold any of that against the study. Those are the reasons to run one.

Where BodyBuddy fits

BodyBuddy works on what happens after the answer. It is an AI health coach that texts you every day. Each day you commit to what you are doing for your health that day, and it checks in to hold you to it, including on the bad days.
  • You text it what you ate, or send a photo, and it logs the meal.
  • If a doctor, dietitian, physical therapist or trainer gave you a plan, you can hand BodyBuddy the plan and it follows up with you between appointments.
  • It checks in by text, so you do not have to remember to open an app.
BodyBuddy is fully AI. There is no human behind the chat, and it does not diagnose anything or replace your doctor.
I think Apple's study app and a coach like ours do different jobs. Apple's helps you understand what is going on and walk into the appointment prepared. Ours texts you the next day to check that you did what you said you would do.

Frequently asked questions

What is the Apple Health AI Usage Study?

It is a research study Apple announced on November 15, 2026, with OpenEvidence. Participants use an iPhone app to ask health questions and get answers with citations to medical research. Apple is measuring trust, the quality of responses, comprehension and everyday health actions.

Who can join the Apple Health AI Usage Study?

Up to 200,000 US adults aged 21 or older can join. You need an Apple Intelligence-enabled iPhone, which means an iPhone 15 Pro or later, running iOS 26 or later. You download the Apple Health AI Study App and complete an informed consent process.

Is Apple's AI health chatbot a doctor?

No. Apple says the app is not meant to diagnose or treat any condition, or replace the advice of medical professionals. ABC reports that it also has safeguards meant to recognize potentially dangerous situations and point you to professional help.

What is OpenEvidence?

OpenEvidence is a clinical AI platform used by healthcare professionals. It answers medical questions using established sources such as The New England Journal of Medicine, JAMA and Nature.

Will Apple use my health data to train AI?

Apple says study data will not be used to train AI models, and that you control whether to share any Health app data or survey responses. ABC reports that OpenEvidence deletes question and response data within 24 hours.

Do AI chatbots give accurate health advice?

On their own, often yes. In the Oxford trial, chatbots identified the right conditions in 94.9% of test cases. People using the same chatbots got there in fewer than 34.5% of cases. The trouble is in the conversation between the person and the model, and that conversation is what Apple's study is built to examine.

What to do with this

If you qualify and you are curious, the study looks like a reasonable thing to join. You get cited answers about your own data, and your ratings feed into research Apple says it will share.
Use it to understand your numbers and to write down better questions for your doctor. Then pick one thing you will do today and tell someone who will ask you about it tonight. If you want that to be a text every day, that is what we built BodyBuddy for.

Sources

Everyone gives you a plan.BodyBuddy texts you until it happens.

BodyBuddy keeps you on track—from meals to doctor's appointments and everything in between.