Do you legally have to tell patients when AI is used?

In a growing number of states, yes, when the AI touches clinical care. California's AB 3030, in effect since January 1, 2025, requires a disclaimer whenever generative AI writes or speaks a patient communication about clinical information, plus clear instructions for reaching a human. Texas, under its Responsible Artificial Intelligence Governance Act effective January 1, 2026, requires providers to disclose when AI is used in diagnosis or treatment. Louisiana requires you to tell patients before you use AI to transcribe a visit. Purely administrative help, like an assistant that books an appointment or gives your hours, generally falls outside these clinical rules, but disclosing anyway is the smart and honest move.

What does California AB 3030 actually require?

AB 3030 applies to health facilities, clinics, physician offices, and group practices that use generative AI to produce written or verbal patient communications about clinical information. Those messages must carry a prominent disclaimer that AI generated them and clear instructions on how to reach a human provider. For written messages the disclaimer goes at the start, for ongoing chats it shows throughout, for audio it is spoken at the beginning and end, and for video it stays on screen. There is one big exemption: if a licensed human reads and reviews the communication before it goes out, the disclaimer is not required.

Does an AI receptionist that books appointments need a disclosure?

If it only handles front desk tasks like scheduling, directions, hours, insurance questions, and taking messages, it is generally outside the clinical communication laws, because it is not generating clinical information or making diagnosis or treatment decisions. That said, transparency is still the right call. Patients trust a practice more when the assistant says up front that it is an automated helper and offers a person on request, and a clear handoff to a human is exactly what these laws are pushing everyone toward. Build disclosure in from day one and you are covered as the rules tighten.

Which AI uses in a practice are the risky ones?

The risk climbs the closer AI gets to clinical judgment. AI that drafts messages about a patient's condition, results, or treatment plan, AI scribes that record and transcribe visits, and any tool that suggests a diagnosis or a treatment are the uses regulators are watching, and several now require disclosure, human review, or patient consent. Front desk automation that never touches clinical content sits at the low end of the risk scale. A simple test: if a human clinician would need to own the decision, AI should assist, not replace, and the patient should know a human is in the loop.

What happens if you do not disclose AI use when the law requires it?

You expose the practice to enforcement by your state medical board or attorney general, potential penalties, and the harder cost, a loss of patient trust if it comes out later that AI was speaking for you without anyone knowing. The 2026 wave of state laws is still young, so most enforcement is just beginning, but the direction is clear. The safe posture is to disclose AI in any clinical communication, keep a licensed human reviewing anything clinical, and always give patients an easy path to a real person. That combination keeps you compliant almost everywhere and keeps trust intact.

Do You Have to Tell Patients AI Is Used?

A call center agent wearing a headset answering a phone, the question every practice now faces about whether AI or a human is on the line — When a patient calls, who is really answering, and do they have a right to know? That question is now written into state law. Photo via Pexels.

A clinic manager asked us a simple question last month that turned out to be anything but simple. She had just turned on an AI helper to answer overflow calls and send appointment texts, and a patient had asked her point blank on the phone: was that a robot that texted me yesterday? She did not know what she was allowed to say, or what she was required to say. She just wanted to do the right thing and not get in trouble.

That worry is everywhere right now. Trade coverage this month, from Medical Economics writing about how AI could change patient care rather than replace it, to Capitol Weekly reporting that the growing use of AI tools has many providers on edge, keeps circling the same nerve. Practices want the help AI gives them. They are just not sure where the guardrails are. So here is the honest, practical answer to the question owners keep asking: do you have to tell patients when AI is involved?

The short answer: it depends on what the AI is doing

The rules that are landing in 2026 do not treat all AI the same, and neither should you. They draw a line between AI that touches clinical care and AI that handles the front desk. Once you see that line, most of the confusion clears up.

If AI is writing or speaking to a patient about their clinical information, their condition, results, a treatment plan, or anything a clinician would normally own, a handful of states now say you must disclose it. If AI is booking an appointment, sharing your hours, or texting a reminder, that is administrative, and it generally sits outside those clinical disclosure laws. Same technology, very different rules, depending on the job it is doing.

3 states (California, Texas, and Louisiana) already require providers to disclose AI in certain patient interactions, and more passed new rules in 2026 (Holland & Knight, state legislation review).

What California already requires (AB 3030)

California moved first and hardest. Its law, Assembly Bill 3030, took effect on January 1, 2025, and it is the clearest template we have for where this is all heading. It applies to health facilities, clinics, physician offices, and group practices that use generative AI to produce written or verbal patient communications about clinical information.

When that happens, two things are required. First, a prominent disclaimer telling the patient the message was generated by AI. Second, clear instructions on how to reach a human provider or staff member. The display rules are specific: for a written message the disclaimer goes at the beginning, for an ongoing chat it stays visible the whole way through, for audio it is spoken at the start and the end, and for video it stays on the screen.

There is one exemption that matters enormously for how you run your practice. If a licensed or certified human reads and reviews the communication before it reaches the patient, you do not need the disclaimer. In plain terms, the law is not trying to ban AI. It is trying to make sure a patient is never left talking to a machine about their health while thinking it is a person, and that a human is always reachable. Keep a clinician in the loop and you satisfy the spirit and the letter of it.

The law is not really about AI. It is about consent and reachability. Patients should know when a machine is speaking for you, and they should always be able to get to a person.

Texas and the 2026 wave

California is not alone anymore. Texas passed the Responsible Artificial Intelligence Governance Act, often shortened to TRAIGA, which took effect on January 1, 2026. For healthcare it requires providers to disclose to a patient, or their representative, when an AI system is used in their diagnosis or treatment. The disclosure has to come before or at the time of the interaction, with a narrow exception for emergencies.

Other states are moving in the same direction. Louisiana now requires a healthcare professional to tell patients ahead of time if they plan to use an AI service to record and transcribe a visit, the AI scribe scenario that so many practices adopted quietly over the past two years. And in its 2026 review of state activity, the law firm Holland & Knight found a second wave of bills across states like California, Colorado, and New York all aimed at the same idea: disclose when a patient communication was generated or heavily drafted by AI, and make sure a human alternative exists.

The pattern is unmistakable. Even if your state has nothing on the books today, the smart practice is to run as if it will next year, because it probably will. Building disclosure and a human handoff in now costs you nothing and future proofs you against a rulebook that is only getting stricter. This is the same reasoning we walk owners through in our guide on what to check before you sign an AI vendor.

Where the front desk sits, and why it is different

Here is the part that should take a lot of pressure off. The most valuable AI for a typical practice is not the clinical kind at all. It is the front desk kind, and that is exactly the kind these laws mostly leave alone.

An assistant that answers the phone after hours, books a new patient, reschedules a visit, shares directions, confirms which insurance you take, or sends a reminder text is not generating clinical information and is not making a diagnosis or treatment decision. It is doing the work a receptionist does. That means it generally falls outside California's clinical communication rule and Texas's diagnosis and treatment rule. This is the same everyday capability we cover in our piece on whether AI agents can book appointments for patients, and it is where the real return on investment lives for most owners.

So do you have to disclose that? Legally, in most places, not yet. But we tell every practice we work with to do it anyway, for one simple reason: trust is your whole business. A patient who finds out later that the warm, helpful voice was automated, and nobody said so, feels tricked. A patient who hears a quick, plain line up front, something like you are speaking with our scheduling assistant and I can get you to a team member any time, feels respected. Disclosure is not just a legal shield here. It is good manners, and it converts better than pretending. We dug into that dynamic in our article on AI transparency and patient trust.

A simple map to keep on your desk

You do not need a compliance department to get this right. You need a way to sort any AI tool into the correct bucket. Ask what the tool touches.

Clinical content, disclose and keep a human on it. Anything that writes or speaks about a patient's condition, results, or treatment, any AI scribe recording a visit, or any tool suggesting a diagnosis. Add the disclaimer, give a clear path to a human, and have a licensed person review clinical output. In California, human review can even remove the disclaimer requirement.
Front desk and logistics, disclosure is optional but wise. Scheduling, reminders, hours, directions, insurance questions, message taking. Not covered by the clinical rules in most states, but say it is an assistant and offer a person anyway. It builds trust and keeps you ready as laws expand.
Always, no exceptions, a human is reachable. Every one of these laws, and every patient's expectation, comes back to this. If a patient wants a real person, they should get one fast. Design that door into every automated touch.

Notice that the safe move and the good business move are the same move. You are not choosing between compliance and care. Doing right by the patient is the compliance.

Where EtherealMinds fits

We build patient acquisition systems for healthcare practices in the United States only, so this is not theory for us, it is how we design every tool we hand a client. Our AI receptionist is built for the front desk on purpose. It answers calls and texts, books and reschedules, and captures the after hours leads that used to go to voicemail and never call back, and it introduces itself as an assistant and hands off to a human the moment a conversation needs one. It stays on the safe side of the line by design, and it keeps a person in reach at all times.

That same principle runs through the whole patient acquisition system we install: AI carries the volume and the speed, people carry the judgment and the trust. When a new law lands, our clients are not scrambling, because their setup already assumed the rules would tighten. If you are adding AI to your practice and you want it to grow your schedule without ever gambling with a patient's trust or your license, that is the exact problem we solve.

The honest takeaway

Do you have to tell patients when AI answers? If the AI is touching their clinical care, in a growing list of states, yes, and you should expect that list to keep growing. If it is handling the front desk, usually not, but you should tell them anyway, because a patient who feels informed is a patient who trusts you. The providers who are on edge about AI right now are mostly worried they will get the line wrong. The line is not hard. Let AI do the volume, keep a human on anything clinical and always within reach, and say plainly when a machine is helping. Do that, and you get every bit of the speed with none of the regret.

Want AI that grows your schedule and keeps you compliant?

Book a free strategy call. We will show you how to put AI on your phones and front desk the safe way, capture more new patients, and stay on the right side of every disclosure rule as the laws keep changing.

Book a free strategy call →

Do You Have to Tell Patients When AI Answers?