OPINION: ChatGPT Takes Your Pulse — Public Health Entrusted to a Language Model

Three Hundred Million Health Questions a Week, Asked to an Algorithm

The Number OpenAI Itself Highlights

OpenAI states, on the page announcing the launch of "Health in ChatGPT" dated July 23, 2026, that "every week, more than 300 million people turn to ChatGPT with health-related questions." That figure, published by the company itself, is not trivial: it signals that ChatGPT was already, before this official launch, a mass health interlocutor — without any professional body or health agency having validated that role.

On that same page, OpenAI announces that "Health in ChatGPT is launching to U.S. users." The shift from massive informal use to an officially named feature changes the nature of the responsibility the company takes on: it no longer merely tolerates a use, it structures it, names it, and encourages it.

From a Tolerated Use to an Officially Owned Product

This semantic shift matters: as long as massive use stays informal, a company can argue it is observing user behavior rather than designing for it. By naming the feature Health and officially launching it, OpenAI takes on a product responsibility it could previously keep at arm's length.

Half a billion health questions a month never asked OpenAI's permission. They got it anyway.

That volume existed long before any press release acknowledged it, which raises a simple question this piece keeps returning to: who was actually deciding how those questions got answered, before anyone called it a product?

Health Connects Apple Health, One Medical, and Function Health

Apple Health, One Medical, Function Health: the Medical Record Enters the Chat

OpenAI's help page, also dated July 23, 2026, specifies that Health lets users connect Apple Health, supported U.S. provider portals, as well as One Medical and Function Health. This is no longer a chatbot answering generic questions: it is a tool that can, with user authorization, ingest real personal health data.

That same page specifies that Health is available to eligible, signed-in ChatGPT Free, Go, Plus, and Pro users in the United States who are 18 or older, and supported on the Web and on iOS. The partial free access — through the Free tier — means the barrier to connecting one's health data to a language model is nearly nonexistent.

Linking your medical record to a free chatbot has never been easier. That's exactly what should worry you.

Ease of use and safety are not the same design goal, and a product built to maximize the first does not automatically deliver the second. Nothing in the pages consulted for this piece suggests OpenAI has resolved that tension publicly.

A Usage History That Goes Back to January, Not July

"ChatGPT for Healthcare" Already Existed, Inside Hospitals

This July 2026 public launch is not OpenAI's first move in the health field. The page "Introducing OpenAI for Healthcare," dated January 8, 2026, states that ChatGPT for Healthcare was "available starting today" and already deployed at AdventHealth, Baylor Scott & White Health, Boston Children's Hospital, Cedars-Sinai Medical Center, HCA Healthcare, Memorial Sloan Kettering Cancer Center, Stanford Medicine Children's Health, and UCSF.

Eight leading American hospital institutions, explicitly named by OpenAI, were already using a professional version of this tool six months before the general public could access it directly through their own personal account. This head start changes how the July launch should be read: it is not a risky technological gamble, it is the public-facing extension of a tool already validated, to some degree, by serious medical institutions.

What Professional Hospital Use Doesn't Guarantee the Public

Clinical use of such a tool, supervised by trained physicians and strict institutional protocols, does not automatically translate to unsupervised individual use. A hospital like UCSF or Cedars-Sinai has internal safeguards — human validation, second opinions, escalation protocols — that no page consulted documents for Health's public-facing use.

Eight leading hospitals opened the door in January. In July, it opened for everyone.

A Figure That Climbed From 230 to 300 Million in Five Weeks

Growth OpenAI Doesn't Fully Explain

The page "Improving health intelligence in ChatGPT," dated June 18, 2026, stated that "every week, more than 230 million people turn to ChatGPT for help with health and wellness questions." Five weeks later, the announced figure jumps to more than 300 million. OpenAI does not specify, in the pages consulted for this piece, whether this rise reflects a real change in usage or a change in counting methodology.

A 30% jump in five weeks, if it reflects real usage rather than a statistical artifact, would represent extraordinarily rapid adoption for a service that touches directly on people's health. Absent a published methodology, this piece limits itself to reporting the two figures as OpenAI published them, without settling their exact comparability.

A number that jumps 30% in five weeks deserves an explanation. It's still missing.

The GPT-5.6 Sol Model, Presented as the Strongest for Health

A Performance Claim, With No Independent Clinical Study Cited

OpenAI presents GPT-5.6 Sol as the model "strongest yet for health" on its launch page. This claim, made by the company selling the product, is not accompanied, in the sources consulted for this piece, by an independent clinical study allowing verification of that superiority relative to other tools or to human health professionals.

The January 2026 "Introducing OpenAI for Healthcare" page instead mentioned GPT-5.2 as the engine behind hospital products of that period. The shift from one model generation to the next, in six months, illustrates how fast this technology is moving — a pace that leaves regulators and independent researchers little time to validate each version before it reaches the public.

"Strongest for health" is a marketing promise. It is not yet a clinical conclusion.

America's Regulatory Gap Demands Almost Nothing of OpenAI

A Health Tool That Isn't a Medical Device

In the United States, software that specifically diagnoses, treats, or prevents a disease generally falls under the FDA's medical device regulation. A tool presented as a general health information assistant, however, can escape that category — and that is precisely the positioning most major conversational AI tools aimed at the general public adopt.

This piece found no mention, in the sources consulted, of a specific FDA authorization or classification for Health in ChatGPT. That absence is not necessarily illegal — it depends on exactly how the product is legally positioned — but it means a tool consulted by hundreds of millions of people every week for health questions operates, as of today, largely outside the regulatory framework designed for medical devices.

Three hundred million users a week, and no medical-device rule applies.

Health Data Privacy, a Question That Remains Unanswered

Connecting Apple Health to a Chatbot, and Then What?

The pages consulted for this piece detail the possible integrations — Apple Health, One Medical, Function Health — but do not detail, in the excerpts available for this piece, the specific protections applied to this data once it enters the ChatGPT ecosystem: retention period, use for training future models, potential sharing with third parties.

In the United States, health data transmitted directly by a patient to a digital service does not automatically benefit from the same protections as data exchanged between health professionals under the HIPAA framework. This piece cannot state with certainty which protection regime exactly applies to data entered into Health via ChatGPT, for lack of detailed documentation available in the sources consulted.

Silence about how long personal health data is kept is not a guarantee. It's a question with no published answer.

The Danish Precedent, Revealing a Global Ambition

230 Million Worldwide, Not Just in the United States

OpenAI's Danish page "Vi præsenterer ChatGPT Health," dated January 7, 2026, states that more than 230 million people worldwide ask ChatGPT health and wellness questions every week. This mention, predating the official U.S. launch by more than six months, reveals that OpenAI's ambition in this field was never limited to the United States.

That same Danish page mentions b.well as a partner enabling access to U.S. health providers — a partner no longer explicitly listed on the American help page of July 23, 2026, which instead mentions One Medical and Function Health. This partnership shift over six months reveals a still-unstable ecosystem, where integrations change faster than public understanding of their implications.

What the Partner Change Signals About the Product's Maturity

Swapping an integration partner within six months isn't necessarily a sign of failure, but it is a sign of a product still under active construction rather than a stabilized architecture. Users who connected their data through a now-discontinued partnership find themselves in a situation whose practical consequences are not detailed in the pages consulted.

OpenAI's ambition in health was global before it was officially American. The public is finding out late.

Three Hundred Million Weekly Questions Betray a Failing Access to Care

A Symptom of Access, Not Just a Technological Feat

A figure of this magnitude doesn't say something just about ChatGPT. It says something about access to care: hundreds of millions of people search every week for answers to health questions from a tool that is free, available at any hour, with no appointment or wait time — three features that many health systems, including American ones, don't consistently offer.

An algorithm doesn't become the world's on-call doctor because it's the best. It becomes that because it's the most available. This distinction, often overlooked in the tech coverage of this launch, deserves to be stated head-on: availability is not the same as clinical competence.

An algorithm available twenty-four hours a day doesn't replace a health system that, itself, isn't.

Liability in Case of Error, a Question No Page Consulted Settles

Who Answers for Bad Health Advice Given by a Language Model?

None of the OpenAI pages consulted for this piece explicitly detail the legal liability regime applicable if ChatGPT gives erroneous health advice to a user who suffers as a result. The general terms of use of large language models generally contain broad liability disclaimers, but their enforceability against a documented health injury remains, to a large extent, largely untested legal ground in the United States.

This legal gray zone is not unique to OpenAI — it concerns the entire conversational AI industry applied to health. But its scale grows mechanically with the number of users: the more people the tool reaches, the higher the statistical probability that erroneous advice causes real harm, even if the model's individual error rate remains low.

What Liability Disclaimers Don't Always Cover

A generic liability disclaimer, buried in terms of use few users read in full, does not necessarily absolve a company of all responsibility in case of documented harm. U.S. courts have not, as of this writing and in the sources consulted for this piece, definitively settled this question for this specific type of product.

Three hundred million users a week turn any error rate, however tiny, into a very real number of harms.

January's Hospital Rollout, a Large-Scale Test Before the Public One

Institutions That Agreed to Be the First Testers

The eight institutions named by OpenAI in January 2026 — including Cedars-Sinai, UCSF, and Memorial Sloan Kettering — rank among the most respected academic medical centers in the United States. Their early adoption of this tool constitutes, in effect, a large-scale test phase whose detailed results do not appear, as of today, in the pages consulted for this piece.

The Silence on Clinical Safety Metrics

None of the pages consulted publishes a clinical safety metric from these hospital deployments — detected error rate, reported incidents, model corrections made following clinical feedback. This silence does not necessarily imply the absence of such incidents; it only implies their absence from OpenAI's public communication to date.

A professional hospital deployment, supervised by trained clinicians and institutional protocols, does not automatically guarantee the safety of a public deployment, where the end user has neither the same training nor the same professional safety net for interpreting the tool's answers.

What works under a clinician's supervision doesn't necessarily work alone, at three in the morning.

The Minimum Age of 18, a Barrier Easy to Get Around

A Declared Restriction, Not Necessarily Verified

OpenAI's help page states that Health is reserved for users aged 18 or older. This restriction, like most age checks on consumer digital platforms, relies largely on a user's self-declaration rather than on rigorous identity verification.

The health questions most sensitive for minors — mental health, sexuality, eating disorders — are precisely those for which the absence of robust age verification poses the highest risk, in a context where this same piece notes that no recommendation touching mental health or a vulnerable population should be made without referral to a qualified professional.

A checkbox doesn't protect a distressed teenager searching for answers at three in the morning.

Medical History Has Already Seen This Kind of New Intermediary

Every Generation Has Had Its Controversial Health Intermediary

The concern raised by a new intermediary between patients and medical information is not unique to artificial intelligence. Search engines, before ChatGPT, had already transformed how hundreds of millions of people sought health information, with the same fears of misinformation, faulty self-diagnosis, and delayed professional consultation.

What sets ChatGPT apart from classic search is the conversational form of the answer: a search engine returns links the user still has to evaluate, while a language model produces a written, confidently toned response that can create an impression of clinical certainty the content doesn't necessarily possess.

A search link invites doubt. A confidently written answer invites trust — sometimes misplaced.

The Tool's Real Effectiveness Remains, to Date, Unproven

No Independent Study Cited in the Available Sources

This piece has no independent clinical study, published in a peer-reviewed journal, allowing an objective evaluation of Health in ChatGPT's answer reliability across a representative sample of real health questions. All the performance claims cited in this piece come from OpenAI's own official pages.

This limitation does not necessarily invalidate OpenAI's claims, but it means no outside evaluation, independent of the company selling the product, is available at this stage to corroborate or contradict them. A careful reader should treat the performance figures cited by the company as marketing claims awaiting external validation.

A company that evaluates its own health product is not proof. It's well-documented advertising.

The Gap This Launch Leaves for Treating Physicians

A Tool That Adds to the Patient-Doctor Relationship, Without Officially Replacing It

None of the pages consulted for this piece claim that Health in ChatGPT replaces a treating physician. But none detail, with precision, how the tool is supposed to fit into a patient's existing medical follow-up either: should they inform their doctor they're using this tool? Are the generated answers designed to be shared with a professional, or only consumed in isolation?

This practical coordination gap leaves each user responsible for deciding, alone, how to integrate a language model's answers into their health decision-making — a cognitive and medical burden historically never entrusted so broadly to non-professionals.

What a Minimal Framework Could Reasonably Require

A minimal framework — a systematic recommendation to share answers with a health professional, a clear warning about the tool's limits for urgent situations, automatic referral to help resources when distress is detected — would not solve every risk, but it would narrow the current gap between the scale of use and the transparency offered to the user.

Nobody explains how a chatbot and a family doctor are supposed to talk. Every patient improvises.

This piece does not call for banning this tool, nor for its moral condemnation. It calls for something more modest and more urgent: that the scale of its use — 300 million questions a week — be matched by equivalent transparency about its limits, its data, its legal liability, and its regulatory framework. For now, scale has outpaced transparency, and it is that gap, more than the technology itself, that deserves attention.

None of this requires villainizing the engineers who built the product, nor the millions of people who use it because nothing better was available to them at two in the morning. It requires naming the gap plainly, in public, so that the next update to these pages closes it rather than simply growing the user count that made it visible in the first place, quietly, one weekly figure at a time, quarter after quarter.

Signed Maxime Marquette, Columnist

Columnist transparency box

Positionnement éditorial

I am not a journalist, but a columnist and analyst. This piece defends an explicit opinion: the scale of Health in ChatGPT's rollout exceeds, as of today, the public transparency available about its limits, its regulation, and the protection of the data it processes.

The columnist is neither a physician nor a health professional. This piece makes no individual diagnosis and does not recommend using or avoiding this tool for any personal case. Any decision touching a symptom, a treatment, or a mental health situation should be made with a qualified health professional.

Méthodologie et sources

This piece distinguishes verified facts and interpretive analysis: the figures, dates, and features cited come exclusively from OpenAI's official pages; the interpretations about regulatory, legal, and privacy risks belong to the columnist.

Primary sources: OpenAI's official pages — the launch announcement, the help page, the page on improving health intelligence, the page on OpenAI for Healthcare, and the Danish introduction page — all consulted directly for this piece.

Secondary sources: no independent third-party media source was identified in the available corpus to corroborate or contradict OpenAI's figures; this absence is itself flagged as a limitation of this piece rather than hidden.

The numerical discrepancies noted in the text — 230 million versus 300 million weekly users, GPT-5.2 versus GPT-5.6 Sol, b.well versus One Medical and Function Health — are explicitly flagged rather than arbitrarily resolved.

Nature de l'analyse

The facts cited (publication dates, usage figures, features, named partners) are reported as published by OpenAI. The analysis of the regulatory gap, legal liability, data privacy, and risks to vulnerable populations belongs to the columnist's interpretation.

Publication of an independent clinical study or an official regulatory classification could substantially change this piece's conclusions.

Sources

Sources primaires et officielles

OpenAI — Launching Health in ChatGPT

OpenAI — Health in ChatGPT (help page)

OpenAI — Improving health intelligence in ChatGPT

Sources secondaires

OpenAI — Introducing OpenAI for Healthcare

OpenAI (Denmark) — Vi præsenterer ChatGPT Health

OpenAI — News