ChatGPT Health: How to Connect Apple Health and What to Know First

Auf July 23, 2026, OpenAI opened ChatGPT Health to every logged-in user in the United States aged 18 or over, on the Free, Go, Plus and Pro plans. It lets ChatGPT read your Apple Health data and your actual medical records, then answer questions about them in plain language. Six months earlier the same feature sat behind a waitlist, which is why most of the advice online about getting access is now out of date.

There is a catch for anyone reading this on a Mac, and a bigger one for anyone thinking of using it in an emergency. Below we cover exactly who can switch it on, how to connect Apple Health from an iPhone, and what OpenAI does with the data. We also cover what an independent study in Nature Medicine found when clinicians stress-tested its triage advice. The short version on that last point is uncomfortable.

The Key Takeaways

• Live since July 23, 2026 for all US users 18 and over, on Free, Go, Plus and Pro. There is no waitlist any more.
• It runs on web and iOS only. It is not in the ChatGPT Mac desktop app, and connecting Apple Health needs an iPhone.
• Still blocked in the EU, the UK and Switzerland over stricter privacy rules, with no announced timeline.
• A Nature Medicine study from Mount Sinai found it under-triaged 52% of the cases physicians graded as genuine emergencies.
• Your records are not covered by HIPAA once they reach OpenAI, because a consumer app is not a covered entity.

What Is ChatGPT Health?

ChatGPT Health is a separate space inside ChatGPT where you can connect Apple Health data and supported medical records, then ask questions about them in ordinary language. It compares a new lab result against your previous ones, summarises what changed since your last appointment, and translates the terminology on a discharge note into something readable. In its launch announcement, OpenAI is explicit that it is built to support medical care rather than replace it, and that it is not intended for diagnosis or treatment.

The reason OpenAI built it is volume. The company says more than 300 million health-related questions reach ChatGPT every week, and that during the earlier pilot roughly 70% of them were asked outside the dedicated health space entirely. People were already using ChatGPT this way, so the July release lets your connected health context inform answers across normal conversations too, not only inside the Health tab.

OpenAI also says it tuned specific models for these conversations, and that GPT-5.6 Luna, the cheapest tier of its newest family, beats GPT-5.5 on HealthBench, its own open-source health benchmark. That is a vendor-run evaluation rather than an independent one, which matters later in this article. If you want the wider context on that model generation, we broke down the whole GPT-5.6 model family when it shipped.

Can You Use ChatGPT Health Right Now?

Four separate things have to line up, and most of the confusion online comes from people quoting the rules of the January pilot instead of the current ones. Here is where things actually stand.

Requirement	What you need	Notes
Country	United States	Blocked in the EU, UK and Switzerland. Everywhere else waits for a wider rollout
Platform	Web or iOS	Not available in the ChatGPT Mac desktop app
Plan	Free, Go, Plus or Pro	No paid upgrade required
Age	18 or over	Account must be logged in
Apple Health	An iPhone	Apple only shares this data through HealthKit on iOS

There Is No ChatGPT Health Waitlist Any More

If you signed up for a waitlist earlier this year, you can stop waiting. The January 2026 version was a gated pilot, and joining it meant putting your name down and waiting for an invitation. The July rollout removed that step for US users completely. Open ChatGPT, look for Health in the sidebar, and if you meet the four conditions above it is simply there.

Where It Is Still Blocked

The European Economic Area, Switzerland and the United Kingdom are excluded, and OpenAI has pointed to the stricter digital privacy regimes in those regions as the reason. No timeline has been announced for any of them. Everywhere else is waiting on a wider rollout, because OpenAI’s own launch wording put this release in the United States and nothing since has extended it. Medical record connections have been US-only from the start, which is unsurprising given the supported systems are all American hospital networks.

ChatGPT Health on Mac, iPhone and the Web

This is the part Apple users keep missing. Health runs on the web and on iOS, and it is not in the ChatGPT desktop app for macOS. If you live in that app, the feature does not exist for you, and the workaround is opening chatgpt.com in Safari or Chrome instead.

The iPhone requirement is a separate constraint and it is Apple’s, not OpenAI’s. Apple Health data is only exposed to other apps through HealthKit, which lives on iOS, so the browser on your Mac cannot reach it no matter which plan you pay for. You can read and discuss data on the web once it is connected, but the connection itself has to be made from the phone.

Practically, an iPhone plus a Mac gives you the full experience, and an iPhone alone gives you almost all of it. A Mac on its own gives you the conversation without any of your own data behind it. We keep a running guide to the best AI apps for iPhone if you are building out the mobile side of this.

How to Connect Apple Health to ChatGPT

You need the ChatGPT app on an iPhone for this, and a few minutes for the first sync to finish.

1. Open the ChatGPT app on your iPhone and find Health in the sidebar. On some accounts it sits under the More menu rather than at the top level.
2. Tap Connect Health or Get started, then choose Apple Health from the list of sources.
3. Work through the permission prompts. Apple will ask which categories you want to share, so this is the moment to be selective rather than tapping allow on everything.
4. Wait for the first sync. It usually takes a few minutes depending on how much history is in your Health app.
5. Review what actually came across before you start asking questions, because gaps in the imported data are a real source of wrong answers.

There is a second route to the same place if the sidebar option does not appear. Go to Health, tap the … menu in the top right corner, then pick Apple Health and choose your data categories from there.

Connecting Medical Records and Wellness Apps

Medical records are a separate connection and they are US-only. In the same Health section you can link a supported provider portal and sign in with the credentials you already use for it. TechCrunch reports that supported systems include Epic, Oracle Health, One Medical und Function Health, which covers a large share of American hospital networks between them.

Wellness apps work the same way. MyFitnessPal und Function are both supported alongside Apple Health, so nutrition and lab-panel data can sit next to your activity and sleep. Each source is connected and disconnected independently, which is worth remembering when you decide how much to hand over.

What Changed Since the January 2026 Pilot

Most stale advice online describes the pilot, so it is worth being precise about the difference between the two releases.

	January 2026 pilot	July 2026 rollout
Access	Waitlist, invitation only	Open to all eligible US users
Where health context applies	Only inside the Health space	Across all your conversations
Platforms	Web and iOS	Web and iOS
Plans	Free, Go, Plus, Pro	Free, Go, Plus, Pro
Medical records	US only	US only

The second row is the substantive change. In the pilot, connected data only informed answers if you were inside the health tab, which is why so many people asked their health questions in a normal chat and got generic responses. Now a question about, say, an allergy can draw on your connected records wherever you happen to ask it.

How Accurate Is ChatGPT Health?

This is where the marketing and the evidence part company. Researchers at the Icahn School of Medicine at Mount Sinai ran a structured stress test of ChatGPT Health’s triage advice and published it in Nature Medicine in February 2026. They used 60 clinician-authored vignettes across 21 clinical domains, varied each one across 16 conditions, and collected 960 responses in total. Urgency levels were graded against guidelines drawn from 56 medical societies.

Among the cases physicians classed as genuine emergencies, the system under-triaged 52% of them. Patients presenting with diabetic ketoacidosis or impending respiratory failure were pointed towards a 24 to 48 hour evaluation rather than an emergency department. Textbook emergencies were handled well, so stroke and anaphylaxis were routed correctly, but the nuanced presentations were where it fell down.

One finding deserves singling out, because it describes something people actually do. When a family member or friend played down the symptoms in the prompt, the model shifted sharply towards less urgent advice, at an odds ratio of 11.7. Reassurance from someone close to you is exactly the thing that makes an emergency sound survivable, and the model inherits that reassurance rather than correcting for it.

Read that as a boundary rather than a verdict on the whole product. Understanding a lab result you already have is a very different task from deciding whether symptoms you are feeling right now need an ambulance. The first is what the tool is good at. The second is what it measurably is not, and no amount of connected data changes that. If you are in a crisis, contact emergency services or a crisis line, not a chatbot.

Is ChatGPT Health Safe? Privacy, Training and HIPAA

What OpenAI Says It Does With Your Data

OpenAI’s commitments here are more specific than its usual privacy language. Connected medical records, Apple Health information, and the conversations that use them are not used to train its foundation models or to target ads. The company says this holds regardless of the model-training setting on your account. It describes purpose-built encryption and isolation designed to keep health conversations compartmentalised from the rest of ChatGPT.

By default ChatGPT asks permission before using connected health data to personalise an answer, and you can either approve each request or switch to always allowing it. ChatGPT does generate memories from your health conversations to inform future chats, though not from the connected records or Apple Health data itself. Temporary Chat is there if you want a one-off question that leaves nothing behind. Our guide on how to use AI without giving up your privacy covers the same controls across other assistants.

Why HIPAA Does Not Follow Your Records

Here is the part that most coverage skips entirely. HIPAA governs covered entities, meaning hospitals, health plans, medical practices and their business associates. A consumer AI product is not one of them. The moment a copy of your record moves from your hospital’s portal into ChatGPT, the federal protection that applied to it inside the hospital stops applying to that copy.

What governs it instead is OpenAI’s terms of service. That is a contract, not a health privacy law. Bradley Malin of Vanderbilt University put it plainly to TIME, describing the arrangement as buyer beware. In a breach you would have no HIPAA-specific rights, though the Federal Trade Commission could act and you could sue directly.

None of that means the current policy is bad, and OpenAI’s stated protections are stronger than the industry norm for consumer apps. It means the protection is a company promise rather than a legal floor, and company promises can be revised in a way that a statute cannot. Decide accordingly, and share selectively rather than connecting everything by reflex.

How to Disconnect and Delete Your Health Data

Leaving is straightforward and worth knowing before you join. You can disconnect any source at any time from Health and then Accounts, and OpenAI says data synced from that source is deleted from its systems within 30 days.

One important exception applies. Anything that already made it into your conversation history stays there until you delete those conversations yourself, so disconnecting Apple Health does not retroactively clean up a chat where you discussed your blood work. For a clean exit, delete the relevant conversations as well as the connection. You can also switch memory off entirely in your settings if you would rather nothing carried forward.

Should You Use ChatGPT Health?

Use it for preparation and comprehension. Understand a result you have already been given. Build a list of questions before an appointment, spot a trend across two years of readings, decode the terminology on a report. On all of those, having your real data attached makes the answers meaningfully better than a generic chat.

Do not use it to decide whether something is urgent. That is the exact task the Mount Sinai researchers measured, and it failed more than half the time on real emergencies. Treat it as something that helps you talk to a doctor rather than something that helps you avoid one. The same caution applies to mental health, where we would point you to our piece on using ChatGPT for your mental wellbeing for the sensible boundaries.

It also costs nothing to try, since it works on the free tier, and you can check what else your plan includes in our guide to whether ChatGPT is free. If the Mac gap is what frustrates you, that is a structural thing rather than a bug, and your health data is going to stay on the phone and the web for now. For the rest of your desktop AI work, Fello AI puts ChatGPT, Claude and Gemini in one native Mac app for $9.99 a month. That is a reasonable home for everything that is not your medical records.

Schlussfolgerung

ChatGPT Health is now free, open to every US adult, and useful for making sense of information you already hold. It is also web and iPhone only, outside HIPAA, and demonstrably unreliable at judging urgency. Those facts sit together comfortably as long as you know all three.

If you want to try it, open ChatGPT on your iPhone, connect Apple Health first and leave your medical records for later. Live with it for a week on questions you already know the answers to, and you will get a fair sense of whether you trust it with the ones you do not.

FAQ