The question clinics ask me is whether AI-written patient material is legal. That is the wrong question. It is legal. The better question is what you will have to disclose, to whom, and how soon — because the direction of travel in 2026 is unambiguous.
What the 2026 laws actually say
State legislatures moved quickly on healthcare AI this year, and four themes dominate: mental health chatbots, patient disclosure and consent, stopping AI tools from presenting themselves as clinicians, and payer use of AI in coverage decisions.
California’s AB 489, effective 1 January 2026, is the clearest example. It prohibits AI tools from misleading patients into believing they are interacting with a human clinician. A practice running an automated intake flow or a symptom chatbot has to say plainly that the user is talking to software.
Read carefully, these statutes are about interaction. They regulate the moment a patient is in a conversation with a system and could reasonably mistake it for a person. That is a narrower target than most of the panicked commentary suggests.
What they do not yet cover
A printed handout is not an interaction. If you generate a leaflet about hypertension with a language model, review it, and hand it to a patient at the end of a consultation, no current state disclosure statute squarely requires you to label it as AI-assisted.
That is the honest answer, and I would rather give it than manufacture an obligation that does not exist. Clinics have enough real compliance burden without imaginary additions.
But the absence of a rule is not the same as the absence of a duty. The material still carries your name, your logo, and the implicit claim that a clinician stands behind it.
Why the gap will close
Regulators rarely stop at the first version of a rule. The disclosure principle now embedded in state law — patients should know when a machine produced what they are receiving — does not have an obvious reason to apply to a chatbot transcript and not to a leaflet.
The FDA’s January 2026 guidance on clinical decision support software pushed in the same direction: the more a tool influences a clinical decision, the more scrutiny attaches to it. Patient education influences decisions. That is the entire point of writing it.
Anyone building a content library today should assume that within a few years they may need to state how each document was produced and who reviewed it. Retrofitting that record across two hundred undated PDFs is miserable. Capturing it as you go costs nothing.
The practical test I apply
Before a document goes out under a clinician’s name, I ask one question: could I defend every clinical claim in this to a colleague who disagreed with me, citing where it came from?
If the answer is yes, the tool that produced the first draft is genuinely uninteresting. A language model, a template, a registrar, a textbook — the accountability sits with the person who signed it either way.
If the answer is no, the document should not go out regardless of how it was written. The failure mode people fear from AI is not new. It is unreviewed content, and clinics were producing that long before generative models existed.
What this means if you buy content
Ask your supplier three things. Who reviewed this, and what are their credentials? What sources support the clinical claims? Will you tell me if your production process changes?
A supplier who cannot answer the first question has sold you a liability rather than an asset. A supplier who answers all three has given you the record you will want if disclosure rules tighten.
This is the practical reason physician review is worth paying for, and it has nothing to do with whether machines can write clearly. They can. It has to do with who is accountable when a patient acts on what the page said.
References used in drafting
Analyses of 2026 state healthcare AI legislation, including California AB 489 effective 1 January 2026, informed the description of disclosure and consent requirements and the four legislative themes.
The FDA’s January 2026 guidance on clinical decision support software informed the discussion of regulatory direction. Nothing here is legal advice; practices should confirm obligations in their own jurisdiction with counsel.
