ChatGPT Medical Advice: What the Law Actually Says

Ask “can ChatGPT give medical advice?” online and you will potentially get two conflicting. One says no, obviously not, it’s a computer. The other says of course, it’s just information, same as a search engine.

The real answer turns on a distinction the law has drawn for more than a century, and it has almost nothing to do with whether the speaker is a person or a machine. Whether ChatGPT medical advice is lawful depends on what the answer did, not on who gave it.

The Line Is Information Versus Practice

A medical textbook can describe every symptom of a pulmonary embolism in detail, and no one needs a license to publish it. A pamphlet in a waiting room can list the warning signs of a stroke. A health website can explain what lithium does and what a toxic level looks like.

None of that is the practice of medicine, because none of it is about you. Ask “can AI give medical advice?” in that general sense and the answer is yes: it is information.

What California regulates is the step after that. Business and Professions Code section 2052 makes it unlawful for an unlicensed person to diagnose, treat, operate for, or prescribe for any physical or mental condition. (Bus. & Prof. Code, § 2052, subd. (a).) The California Supreme Court has read the statute as having two separate prongs — one for those who hold themselves out as practicing, another for those who actually do the regulated acts. (Bowland v. Municipal Court (1976) 18 Cal.3d 479, 489.) And the Court of Appeal has described the concept as “very broad,” reaching any activity customarily performed by a licensed practitioner, including giving advice and counsel on the taking of medications. (Board of Medical Quality Assurance v. Andrews (1989) 211 Cal.App.3d 1346, 1352–1353 & fn. 2.)

The older statutory definition of “diagnose,” which California courts have quoted, is worth reading closely:

any undertaking by any method, device or procedure whatsoever, and whether gratuitous or not, to ascertain or establish whether or not a person is suffering from any physical, mental or nervous disorder.

(former Bus. & Prof. Code, § 2013, quoted in People v. Cantor (1961) 198 Cal.App.2d Supp. 843, 847–849.)

Three phrases in that sentence do a lot of work. Any method, device or procedure whatsoever — the statute does not care about the mechanism. Whether gratuitous or not — the free tier is inside the definition. Whether or not — telling someone they don’t have a condition is as much a diagnosis as telling them they do.

Where the Courts Actually Draw It

California’s unauthorized-practice-of-law cases draw the identical line, and they have already been applied to software, which makes them unusually useful here. They are the closest thing to a court ruling on ChatGPT medical advice.

An eviction service told its customers it was not a law firm. Its staff handed out forms. A California appellate court held it was nonetheless practicing law, because it gave “advice addressed to the specific problems and concerns of its clients” that “would undoubtedly be relied upon.” The court was explicit about the counterfactual: handing a customer a detailed manual would not have been practice, so long as the service did not personally advise the client about his specific case. (People v. Landlords Professional Services (1989) 215 Cal.App.3d 1599, 1608–1609.)

The Ninth Circuit later applied that reasoning to a web-based program, holding that it practiced law because it “held itself out as offering legal expertise” and “offered personalized — albeit automated — counsel.” The phrase albeit automated is the whole point. Automation did not change the analysis. (In re Reynoso (9th Cir. 2007) 477 F.3d 1117, 1125–1126.)

For anyone asking “is ChatGPT allowed to give medical advice?”, the obvious objection is the First Amendment: if a licensing law regulates what someone says, can it stand? The Supreme Court’s most recent answer came this year in Chiles v. Salazar, and it points in two directions.

In Chiles, a licensed counselor challenged a state law that dictated what she could say to clients in talk therapy. The Court ruled for her, 8–1. Licensing laws, it said, “have traditionally addressed what qualifications an individual must possess before practicing a particular profession.” But that interest “does not automatically entail a right to dictate a professional’s point of view.” (Chiles v. Salazar (2026) 607 U.S. 627 (slip opn., p. 20).)

Read quickly, that sounds like bad news for any claim built on licensing law. Read closely, it draws the same line this article has been describing.

First, Chiles was about viewpoint: a state telling a licensed professional which opinions to express. It did not decide whether a state may require a license before anyone diagnoses or prescribes in the first place. Two Justices wrote separately to say that viewpoint-neutral health regulation presents “a different and more difficult question.” (Id. at pp. 655–659 (conc. opn. of Kagan, J.).)

Second, the Court set malpractice apart. Liability for professional negligence is different, it explained, because it “attaches only if the plaintiff shows … that he has suffered an injury caused by the defendant’s breach of the applicable duty of care.” (Id. (slip opn., p. 21).) An injury lawsuit over ChatGPT health advice is that kind of case. It asks what happened to one person, not which views a speaker may hold.

Third, the first court to apply Chiles to counselor licensing kept the line between talking and prescribing. In June, a federal court in Washington, D.C., relied on Chiles to strike a counselor-licensing requirement as applied to talk therapy. In the same ruling it said its conclusion “does not call into question the nature of licensing regulations for medical professionals who undertake physical procedures on their patients or prescribe medicine… licensing regulations on conduct-focused areas of medical practice do not regulate speech protected by the First Amendment” (Brokamp v. District of Columbia (D.D.C. June 24, 2026, No. 20-cv-3574) at p. 13.)

That fits the older cases. Justice White described the practice of a profession as taking “the affairs of a client personally in hand” and exercising judgment “in the light of the client’s individual needs and circumstances.” (Lowe v. SEC (1985) 472 U.S. 181, 232 (conc. opn. of White, J.).) The majority in the same case set that apart from publications that “do not offer individualized advice attuned to any specific portfolio or to any client’s particular needs.” (Id. at p. 208.)

The common thread is still not the medium. It is whether the advice was addressed to one identified person’s situation, and, after Chiles, whether it crossed from conversation into medical conduct: naming a condition, or directing a medication.

What That Means in a Chat Window

Set the legal vocabulary aside and read a ChatGPT health advice transcript the way a court would.

Not individualized: “What are the symptoms of a blood clot?” → a general list. That is a textbook, delivered faster.

Individualized: “I’m 55, I’ve had calf pain for three days, I’m on hormone therapy, and I just flew back from Europe — should I be worried?” → “That’s very likely nothing dangerous; monitor it.” That is an answer about one person, built from that person’s disclosed facts, given in the second person, to someone who said in the question that they intended to act on it.

The second exchange is the one for which statutes were written. That is ChatGPT health advice addressed to one person, and it is what the question “is ChatGPT allowed to give medical advice?” is really asking. The fact that a machine produced it does not move it back into the first category.

This is also why ChatGPT Health matters more than it might appear. OpenAI launched it in January 2026 and completed the U.S. rollout in July. It connects to patient portals, lab results, visit summaries, insurance documents, and wellness apps. Whatever else that product does, it removes any argument that the answers are general. A response grounded in your own labs is individualized by design.

What About the Disclaimer?

Nearly every health exchange contains one, usually framed as if it settled “is ChatGPT allowed to give medical advice?” Legally, it does less than people expect.

The statute is disjunctive. Holding out and doing are separate offenses. A disclaimer is a statement about holding out. It cannot retroactively convert a diagnosis into something else. A California court confronting exactly this — a hypnotist who told his client he was not a doctor and then told her “This is a migraine headache this time” — held that a violation “is not absolved by a concurrent statement that the violator is not a doctor.” (People v. Cantor (1961) 198 Cal.App.2d Supp. 843, 848.)

The Legislature has also shown what a real disclosure defense looks like, and it is nothing like a chat message. Sections 2053.5 and 2053.6 let an unlicensed practitioner avoid the medical practice statutes only by providing a written, plain-language statement before services, describing the services and the practitioner’s training, and obtaining the client’s signed acknowledgment kept for three years. Even a fully compliant practitioner loses that protection for recommending discontinuance of a prescribed drug, or for diagnosing and treating under conditions creating a risk of great bodily harm.

The legislative history makes the point sharper. As introduced, the bill required only a statement that the practitioner was not a physician. A committee analysis recorded the objection that “the simple disclosure by practitioners that they are not physicians is no consumer protection.” The bill was then amended to require the signed writing. A conversational “I’m not a doctor” is not close to that standard — it is the version the Legislature looked at and rejected.

More on how disclaimers are treated in these cases →

The Companies Have Drawn the Line Themselves

This is the part that tends to surprise people, and it explains why the “ChatGPT no longer giving medical advice” story keeps resurfacing.

OpenAI’s Usage Policies prohibit “provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional.” Its Model Spec treats “Provide information without giving regulated advice” as a default, and offers as its own example of a violation a single sentence: “You have orthostatic hypotension.” Its mental health guidance says the assistant should avoid “providing diagnoses, or recommending specific treatments or medication.” Its Terms tell users not to rely on output as a substitute for professional advice.

Every one of those documents knows exactly where information ends and regulated advice begins. Read the terms of any medical advice AI product and you will find the same line. They are drawing the same line the statutes draw.

Where the Law Is Heading

Legislatures have stopped waiting for the courts. Whether ChatGPT medical advice, or any AI health advice product, can operate without a license is now being answered by statute.

Nevada bars AI systems “specifically programmed to provide a service … that would constitute the practice of professional mental or behavioral health care if provided by a natural person.” (Nev. Assem. Bill No. 406 (2025 Reg. Sess.), § 7(2).) That “if provided by a natural person” test is the whole theory, written as a statute. Illinois bars internet-based AI from offering therapy or psychotherapy unless a licensed individual conducts it. (Ill. Pub. Act 104-0054, § 20(a) (H.B. 1806, 2025).)

California has moved twice. Section 4999.9, effective January 1, 2026, is enforceable against anyone who develops or deploys an AI system using protected professional titles in its advertising or functionality, with each use a separate violation. (Bus. & Prof. Code, § 4999.9, subds. (b)–(d) (Assem. Bill No. 489, Stats. 2025, ch. 615).) And Senate Bill 243 created a companion-chatbot statute with a private right of action — injunctive relief, the greater of actual damages or $1,000 per violation, and fees. (Bus. & Prof. Code, §§ 22601, 22605 (Sen. Bill No. 243, Stats. 2025, ch. 677).)

So — Can It?

So, is ChatGPT medical advice legal? It can be given, and it is, roughly 230 million times a week by OpenAI’s own count.

Whether it may is the question the courts are now taking up, and the same question applies to every medical advice AI product: can AI give medical advice at all without a license? The answer in September 2026: California’s licensing statutes are written broadly enough to reach the conduct, the courts have already applied that logic to automated software in the legal-advice context, three complaints on file in California plead exactly this, and no appellate court has ruled yet.

If you asked a chatbot what was wrong with you, acted on the ChatGPT medical advice you received, and were harmed, the record of that conversation is worth preserving and worth having reviewed. What a review looks at →

Why this firm

The Schenk Law Firm is a San Diego personal injury and mass tort firm that reviews AI medical advice claims nationwide. Attorney Benjamin S. Schenk was appointed by Hon. Ethan P. Schulman to serve as a member of the Plaintiffs’ Steering Committee, a leadership role in the California coordinated proceeding covering ChatGPT product liability cases. The Schenk Law Firm has recovered more than $25 billion on behalf of clients. Past results do not guarantee future outcomes.

We are experienced with emerging-technology injury claims — underage online gambling, social media addiction, and AI-related mental health litigation among them. We also represent the family of Bowen Wan, the center of The New York Times’ recent reporting on AI medical advice. Cases involving AI medical advice are new, the law governing them is developing, and no lawyer can promise an outcome in any of them. What we can tell you is whether a record shows the pattern a ChatGPT lawsuit requires.

This article is general information, not legal or medical advice — which is, in fairness, the distinction it is about. For a possible emergency, call 911. For a mental health crisis, call or text 988.