ChatGPT Health: OpenAI launches its AI health assistant

230 million weekly users were already asking ChatGPT health-related questions before OpenAI decided to make it official.

On January 7, 2026, the company launched ChatGPT Health: a dedicated, compartmentalized space designed to handle medical data with a level of security that standard ChatGPT could not guarantee.

This is not just another chatbot that answers “see your doctor” to every question.

This is a system connected to your electronic health records, your wellness apps, and capable of analyzing your lab results in a natural conversation.

Here is everything you need to know about this new piece of the OpenAI ecosystem, its actual capabilities, its limitations, and what it means for Europe.

What is ChatGPT Health?

ChatGPT Health is not a separate application.

It is a dedicated space inside ChatGPT, isolated from the rest of your conversations.

This architectural separation serves a clear purpose: handling sensitive health data without mixing it with your everyday chats about chocolate cake recipes or your next marketing strategy.

A compartmentalized architecture for sensitive data

The system relies on asymmetric memory.

In practice, ChatGPT Health can access context from your general conversations to better understand your situation, but the reverse is impossible.

Your medical data stays locked inside the Health space and never “leaks” into standard ChatGPT.

ChatGPT Health’s asymmetric architecture is unprecedented in the industry: your medical data is siloed, but the AI retains the context it needs to deliver relevant answers.

This compartmentalization comes with custom instructions specific to the Health space.

You can specify your medical history, allergies, health goals, and ChatGPT Health stores them separately.

Available modes cover a wide spectrum: search, deep research, voice mode, dictation, and even photo and file uploads (medical reports, imaging results).

Who can access it? (Plus, Pro, Team, Enterprise)

ChatGPT Health is not free.

Access is limited to Plus, Pro, Team, and Enterprise subscribers.

The rollout started in the United States with a waitlist system, before a gradual expansion to other regions.

For Enterprise and Edu subscribers, integration goes through OpenAI’s Apps SDK, a technical layer that opens up deeper integration possibilities in professional environments.

Key features of ChatGPT Health

The value of ChatGPT Health goes far beyond asking an AI medical questions.

Its strength lies in its ability to connect to real data sources and ground its answers in your personal situation.

Connection to medical records via b.well

The partnership with b.well Connected Health is the cornerstone of the offering.

b.well aggregates data from 2.2 million healthcare providers and over 320 insurance plans in the United States.

The data accessible through this connection is detailed: lab results, imaging reports, diagnoses, prescribed medications, immunizations, clinical notes, and care plans.

With 2.2 million connected providers, b.well turns ChatGPT Health into a true translator of medical records that are often unreadable for patients.

The idea is not for OpenAI to store your medical records on its servers.

b.well normalizes data into the FHIR (Fast Healthcare Interoperability Resources) format, a medical interoperability standard, before passing it to the AI for interpretation.

Imagine asking “How has my HbA1c level trended over the past 12 months?” and getting a contextualized answer based on your actual results, instead of a generic text about diabetes.

Integrated wellness apps (Apple Health, Peloton, MyFitnessPal…)

Beyond medical records, ChatGPT Health connects to a wide range of wellness applications.

Here is what each integration brings:

Application	Data transmitted
Apple Health	Movement, sleep, physical activity
Function	Lab tests, nutrition
MyFitnessPal	Nutrition, macros, recipes
Weight Watchers	Meals, GLP-1 programs
Peloton	Exercise classes, meditation
AllTrails	Hikes, trails
Instacart	Grocery lists

The Instacart integration deserves a closer look.

It closes the loop between diagnosis (“your cholesterol is high”), nutritional advice (“focus on omega-3s and fiber”), and concrete action (“here is your tailored grocery list”).

This complete chain is what sets ChatGPT Health apart from a simple medical search engine.

Personalized analysis and conversational mode

ChatGPT Health does not just display your data.

It interprets it in plain language.

An imaging report packed with radiology jargon becomes a clear explanation of what the doctor observed.

A blood panel with 40 lines of values turns into an actionable summary.

Voice mode opens an unexpected dimension: you can literally discuss your medical results with the AI the way you would with a friend who happens to be a doctor.

Conversational voice mode adds another layer to the experience.

You can discuss your results out loud, ask follow-up questions, and get clarifications without ever touching a keyboard.

For elderly people or those who are not comfortable with written communication, it provides unprecedented access to understanding their own health.

How OpenAI developed ChatGPT Health

OpenAI did not launch this product on a whim.

The development involved a medical validation process rarely seen in the tech industry.

260 doctors, 60 countries, 600,000 evaluations

Over 260 doctors from 60 countries participated in training and evaluating the system.

These professionals span 30 medical specialties, from general practitioners to oncologists.

In total, 600,000 feedback entries were collected on the model’s responses, making it one of the most extensive peer review processes ever conducted on an AI product.

This groundwork explains why ChatGPT Health knows when to rephrase an overly technical answer and when to explicitly refer users to a specialist.

The AI learned the clinical nuances that only experienced practitioners can convey.

HealthBench: an unprecedented evaluation framework

OpenAI created HealthBench, a health-specific evaluation framework.

This framework includes 48,562 clinical rubric criteria organized around 7 key themes.

This is not a simple true/false test on medical questions.

HealthBench evaluates answer quality across dimensions like clinical accuracy, communication clarity, recommendation relevance, and the ability to identify emergency situations.

With 48,562 evaluation criteria, HealthBench sets a standard that competitors will need to match if they want to be taken seriously in AI healthcare.

This framework is public, meaning other players in the field can use it to evaluate their own models.

A strategic move by OpenAI to set its standards in a market where the race for performance never stops.

Privacy and data security

This is THE question that comes up with every AI announcement touching on health.

And OpenAI seems to have taken it seriously, at least on paper.

Data not used for training

Health data shared in ChatGPT Health is not used to train OpenAI’s models.

This is an explicit commitment, different from the standard ChatGPT policy where conversations can contribute to model improvement (unless you opt out).

On b.well‘s side, the infrastructure is HIPAA, SOC 2, and HITRUST certified, three recognized security standards in the American healthcare sector.

Data is encrypted and travels through secure channels before being presented to the AI.

Explicit consent and user control

Every connection to an app or medical record requires explicit consent from the user.

No pre-checked boxes, no silent opt-ins.

Users retain full control over which data is shared and can revoke access at any time.

b.well complies with the CARIN Code of Conduct and holds the DiMe Seal, two certifications that govern health data sharing in the United States.

One important detail: even though your data is not used for training, it is temporarily processed by OpenAI’s servers to generate a response.

The distinction is subtle, and cybersecurity experts consider it insufficient to guarantee absolute confidentiality.

Limitations and controversies

Like any AI product, ChatGPT Health comes with limitations that would be irresponsible to ignore.

ChatGPT Health does not replace a doctor

A study published in PubMed reveals a concerning figure: ChatGPT Health fails to flag more than 50% of medical emergencies.

Across 960 clinical scenarios covering 21 specialties, the system correctly identified urgency in only 48% of genuinely urgent cases.

For non-urgent situations, the rate of appropriate responses drops to 35%.

A system that misses one in two medical emergencies raises a serious concern, no matter how many doctors helped build it.

These numbers are a stark reminder: ChatGPT Health is a tool for understanding and support, not a substitute for medical diagnosis.

OpenAI states this clearly in its terms of use, but the temptation for users to self-diagnose remains strong.

Limited availability (US first, then global)

At launch, ChatGPT Health is only available in the United States.

The connection to medical records via b.well depends on the American healthcare system and its interoperability standards.

For the rest of the world, equivalent partnerships with local health data aggregators will need to be established first.

In France, the Dossier Medical Partage (DMP) and Mon Espace Sante are potential equivalents, but the regulatory constraints imposed by the AI Act and CNIL add extra layers of complexity.

GDPR and Europe: when will France get access?

The arrival of ChatGPT Health in Europe faces a solid regulatory wall.

The GDPR classifies health data as sensitive data, subject to strict processing rules.

Transferring medical data to American servers raises a major legal issue since the European Court of Justice invalidated the Privacy Shield.

Even with the current Data Privacy Framework, the question of data localization remains a powder keg.

OpenAI would need to set up servers in Europe for health data processing.

The company would also need to obtain specific certifications from national authorities and potentially redesign its architecture so that medical data never leaves European territory.

No date has been announced for a European launch.

What this changes for the future of AI in healthcare

The launch of ChatGPT Health marks a turning point: not a technical one, but a strategic one.

For the first time, a consumer-facing AI player is integrating medical data directly into its offering.

The race between OpenAI, Google and specialized players

Google has been working on Med-PaLM since 2023 and has integrated health features into Gemini.

Microsoft, OpenAI’s longtime partner, is pushing forward with Nuance DAX Copilot on the healthcare professional side.

Anthropic already offers Claude for Healthcare to hospitals — and beyond healthcare, a growing number of users are switching from ChatGPT to Claude for their daily AI needs.

What sets ChatGPT Health apart is its consumer-facing positioning.

While others target healthcare professionals, OpenAI is going straight to the end patient: the person who wants to understand their blood work on a Sunday night without waiting until Monday to call their doctor.

With 40 million daily health queries already logged on ChatGPT, OpenAI holds a massive competitive advantage: the user base.

The strategy mirrors the approach taken with Frontier AI agents for enterprise: staking a claim on a vertical market with a product already adopted by millions.

Toward personalized AI health for everyone?

The long-term promise of ChatGPT Health goes beyond browsing medical records.

OpenAI is building a system that could eventually cross-reference your physical activity data (Apple Health), your diet (MyFitnessPal), your lab results (Function), and your prescriptions.

The goal: a holistic view of your health.

This data cross-referencing opens up concrete possibilities: early trend detection, drug interaction alerts, nutritional recommendations based on your actual biomarkers.

The flip side is the risk of excessive dependency on a private system for managing your health.

If OpenAI changes its terms of service, raises its prices, or alters its partnerships, millions of people could lose access to a tool that has become central to their health monitoring.

The question of data portability and interoperability across platforms will come up sooner or later.

Conclusion

ChatGPT Health is not just another feature.

It is a strategic bet by OpenAI on a digital health market valued at $660 billion by 2028.

The product has real strengths: deep integration with medical records, a rigorous medical validation process involving 260 doctors, and an evaluation framework (HealthBench) that could become the industry benchmark.

Its weaknesses are just as real: a concerning failure rate on emergencies, US-only availability, and privacy questions that have not all been satisfactorily answered.

For web professionals and tech entrepreneurs, the signal is clear: AI is moving into healthcare, and the opportunities (and responsibilities) that come with it deserve serious attention.

FAQ

Is ChatGPT Health free?

No, ChatGPT Health is limited to Plus ($20/month), Pro ($200/month), Team, and Enterprise subscribers.

Is ChatGPT Health available in France?

Not yet: the service launched in the United States in January 2026, and no date has been announced for Europe, mainly due to GDPR constraints.

Is my medical data used to train OpenAI’s models?

No, OpenAI commits to not using health data shared in ChatGPT Health for model training.

What medical records can be connected to ChatGPT Health?

Through b.well, you can access your lab results, imaging reports, diagnoses, medications, immunizations, clinical notes, and care plans.

Can ChatGPT Health replace a medical consultation?

No: studies show the system fails to identify more than 50% of medical emergencies, and it should be used as a comprehension tool, not as a substitute for medical diagnosis.

Which wellness apps are compatible?

Apple Health, MyFitnessPal, Peloton, AllTrails, Function, Weight Watchers, and Instacart are among the integrations available at launch.

How does ChatGPT Health’s memory work?

The memory is asymmetric: ChatGPT Health can read context from your general conversations, but your general conversations cannot access your health data.

How many doctors participated in the development?

Over 260 doctors from 60 countries contributed, generating 600,000 evaluations across 30 medical specialties.

What is HealthBench?

HealthBench is the evaluation framework created by OpenAI to test the AI’s medical responses, featuring 48,562 clinical criteria organized around 7 key themes.

Does ChatGPT Health pose a privacy risk?

Despite OpenAI‘s commitments and b.well‘s certifications (HIPAA, SOC 2, HITRUST), experts point out that data is temporarily processed by OpenAI‘s servers, which does not guarantee absolute confidentiality.