How to Set Up ChatGPT Health and Connect Medical Records on iOS

Photo by Mindfulness Com on Unsplash

By the time you finish this guide, ChatGPT Health will be reading your actual lab values, your last visit summary, and your sleep patterns from Apple Health, and answering questions grounded in that data instead of generic web text. The setup itself takes under fifteen minutes. The decisions you make along the way, particularly around what records to connect and whether to enable multi-factor authentication, matter far more than the clicks. This walkthrough explains how to use ChatGPT Health from first launch through verifying your medical records have synced, with the privacy trade-offs called out where they actually exist.

A quick reality check before you start. ChatGPT Health is for personal wellness and HIPAA does not apply to consumer health products like ChatGPT Health. OpenAI's head of health Nate Gross confirmed as much to Medical Economics, explaining that HIPAA covers clinical settings, not consumer apps. If you need HIPAA protections, you need a clinician using ChatGPT for Healthcare, which is a separate product.

Check that your account and region qualify

Access is gated by both subscription tier and geography. Users can connect medical records from U.S. healthcare providers through a partnership with b.well, as well as wellness apps such as Apple Health, Function and MyFitnessPal. Medical record integrations and some app connections are currently available only in the U.S., and Apple Health requires iOS.

Eligibility, per OpenAI's launch page:

• Plans supported: Free, Go, Plus, and Pro
• Regions excluded from early access: European Economic Area, Switzerland, and the United Kingdom
• Apple Health sync: iOS only, regardless of plan
• Medical records via b.well: U.S. providers only at launch

If you sit outside the U.S., you can still use the Health space for general wellness questions and Apple Health data on iOS, but the b.well medical-record connector will not appear. Rest of World reported that OpenAI plans to expand into markets including India, Brazil, Mexico, and the Philippines, though no firm dates have been announced.

Open the Health space and run the first-time setup

The Health area is not a separate app. It lives inside the regular ChatGPT client as its own tab.

1. Update the ChatGPT app to the latest version, or open chatgpt.com on the web.
2. In the left sidebar, look for Health as a distinct entry below your normal chat list. According to Medical Economics, Health appears as a separate tab with its own chat history and memories.
3. Tap or click Health. You will see a one-time intro screen explaining that this space is sandboxed.
4. Accept the Health-specific terms. These are different from your standard ChatGPT terms because of the data-handling rules that apply only to this tab.

Once you are in, anything you type or upload here stays in here. Your Health conversations, connected apps, memory, and files are available only within Health. Health information does not flow back into your main chats. If you ask your usual ChatGPT about a recipe an hour later, it will not know your cholesterol numbers.

Turn on multi-factor authentication before you connect anything

Do this step now, not later. Once medical records are flowing into the account, the cost of a credential leak rises sharply. OpenAI lets you enable MFA from your standard account security settings, and the company explicitly recommends it for Health users. You can further strengthen access controls by enabling multi-factor authentication (MFA), which adds an extra layer of protection to help prevent unauthorized access.

Use an authenticator app rather than SMS where possible. SIM-swap attacks remain the weak point of phone-based 2FA, and a clinical history is a far more sensitive payload than a social media account.

Connect Apple Health for daily metrics

Apple Health is the easiest source to add and gives the Health model the most continuous data. The connector pulls movement, sleep, and activity, which is what makes the assistant useful for questions about patterns rather than one-off symptoms.

On an iPhone running a current iOS:

1. Inside the Health tab, tap the + icon in the composer or open Settings → Apps.
2. Choose Apple Health.
3. iOS will surface its native HealthKit permission sheet. You can allow categories individually. Granting only sleep and activity, for example, is a reasonable middle ground if you do not want menstrual cycle or mental-health entries shared.
4. Confirm and return to ChatGPT. The first sync usually completes in under a minute for typical histories.

If the Apple Health option does not appear, you are almost certainly on Android or web. The integration is iOS-exclusive at launch, per OpenAI.

Connect U.S. medical records through b.well

This is the part most people came for. OpenAI partnered with the digital health company b.well to handle the actual provider connections. B.well operates a health data network designed to support consumer-mediated access to live clinical data across more than 2.2 million providers and 320 health plans, labs and other sources. Built on FHIR-based application programming interfaces (APIs) and trusted healthcare exchange frameworks, the network maintains strict controls around identity verification, consent management, data security and auditability, according to the company.

In the Health tab:

1. Open Apps in Settings, or tap the tools menu in a Health chat.
2. Select Medical Records.
3. You will be redirected into a b.well-hosted flow. Search for your hospital, clinic, lab, or insurer by name. The b.well network covers 2.2 million U.S. providers per Medical Economics, so most major systems including Epic MyChart and Cerner-backed portals are reachable.
4. Authenticate with the same credentials you use for that provider's patient portal.
5. Approve the data categories b.well requests. These typically include lab results, visit summaries, medications, and clinical history.
6. Return to ChatGPT. Initial ingestion can take several minutes for accounts with years of history because b.well runs the data through a cleaning pipeline before it reaches the model.

Repeat for each provider. If you split care between, say, a primary care group, a specialist network, and an independent lab, you will want all three connected for the assistant to see the full picture.

One important caveat on disconnection. Disconnecting can also delete imported data for certain apps (for example, disconnecting Medical Records deletes your records from our partner b.well). Disconnecting does not automatically remove information already referenced in your Health chats or Health memories; delete those conversations and/or memories to remove that content from ChatGPT. If you ever want to fully revoke access, you have to do both: disconnect the app and clear the relevant Health chats.

Add wellness apps that match how you already track

Only add what you actually use. Each connected source widens the surface area for things to leak or get confused. The current supported list, per Medical Economics, includes MyFitnessPal, Weight Watchers, Peloton, AllTrails, Instacart, and Function.

Function is the more interesting addition for anyone who has paid for a longitudinal lab panel outside of insurance. It feeds biomarker trends back into the conversation, which is something even your provider's portal often cannot do across multiple draws.

To add any of these, open Apps in Settings, find the service, and authenticate. The flow is standard OAuth.

Verify the model is actually using your data

Fresh installs sometimes fail to surface connected data on the first prompt. Run a quick check:

• Ask: "Summarize my most recent lab results and flag anything outside the reference range." If records are connected and synced, the response will quote specific values.
• Ask: "What was my average sleep duration last week?" Apple Health should answer with a numeric mean.
• Ask: "List the medications currently in my chart." b.well-pulled records should populate this.

If any of these come back generic, the connection has not finished or permissions were narrower than you intended. Reopen the relevant app in Settings and confirm scopes.

Set custom instructions specific to Health

The Health tab supports its own custom instructions, separate from your main ChatGPT profile. This is where you tell the model whether you want plain-language explanations or clinical terminology, what reading level to target, and whether to suggest follow-up questions to bring to a doctor. Set these once and every Health chat inherits them.

A practical configuration: ask the model to always cite the specific lab value or visit note it is drawing from, and to flag when it is reasoning from general medical knowledge versus your records. This makes hallucinations easier to catch.

Understand what you are trading for the convenience

Convenience has a cost, and OpenAI has been more candid about it than most consumer launches. Conversations in Health are not used to train our foundation models. That is the headline protection. The harder caveat, reported by Medical Economics, is that OpenAI acknowledges health data could still be obtained via subpoena or court order.

The strategic question is whether the value of an AI that reads your full chart outweighs the legal exposure of having that chart sit on a third party's servers. For most people asking about cholesterol trends or preparing for an appointment, yes. For anyone whose health data is potentially evidence in a custody case, employment dispute, or jurisdiction with hostile reproductive-care laws, the answer is more complicated. Rest of World noted that researchers, ethicists, and medical professionals have warned of risks from AI bias and hallucinations in health contexts, and those concerns compound when the model has more personal data to anchor its mistakes to.

There is also a less-discussed trade-off worth flagging. The b.well pipeline cleans and standardizes records before the model sees them. That means if your provider mis-coded a diagnosis or your lab used a non-standard reference range, those errors will be reflected in every Health response until you correct them at the source. The AI is not auditing your chart for accuracy. It is reading what is there.

Why the launch matters beyond the feature list

Apple Health Records, Google Health, and Microsoft HealthVault all tried to centralize personal medical data. None achieved mass adoption. Emarketer analyst Rajiv Leventhal, speaking to Fierce Healthcare, argued that ChatGPT Health's differentiator is the installed base ChatGPT already has, while prior efforts stalled on fragmented access and privacy concerns.

The scale numbers explain the bet. Health-related inquiries are already among the most common uses of ChatGPT, with more than 230 million people globally asking health and wellness questions each week, according to OpenAI's de-identified analysis of user conversations. Per Fierce Healthcare, OpenAI worked with more than 260 physicians across 60 countries over two years, and those clinicians scored model outputs more than 600,000 times across 30 health domains. That is not a guarantee of accuracy, but it is more clinical input than any prior consumer health AI has disclosed.

For more on how OpenAI's broader product line is evolving alongside this, see our coverage in the AI Tools articles and Health Tech articles sections, including the recent GPT-5.5 release which shares safety infrastructure with the Health model. Anyone weighing the privacy trade-offs may also find context in our Privacy & Security articles.

Set up the Health space, connect what you actually use, enable MFA, and treat the assistant as preparation for medical conversations rather than a substitute for them. That is the use case OpenAI built it for. That is also the use case where it earns the data you give it.