Key takeaways
- Yes. Disclose AI in every channel, in the first sentence or line, with a clear way to reach a person.
- Trust is fragile: six in 10 US adults would be uncomfortable with a provider relying on AI for diagnosis and treatment[1], so surprises cost more than disclosure.
- Some states already require it. California requires disclaimers on AI-generated clinical communications unless a licensed provider reviews them[2], and Texas requires providers to disclose AI used in health care services[3].
- Disclosure does not widen what the AI may do. It still stays out of clinical judgment and still needs a human route.
The short answer is yes
Every practice using AI to talk to patients should say so, plainly and early. The ethical case is simple: patients are entitled to know whether a person or software is handling their request, especially in health care, where they may share sensitive information. The practical case is just as strong. A patient who discovers halfway through a call that the friendly voice was automated feels misled, and that feeling attaches to the practice, not the software vendor.
There is also a growing legal case, covered below and in more detail in state laws on AI in patient communications. Practices that disclose by default rarely need to track which rule applies to which message. Practices that disclose only when required end up maintaining a complicated map of exceptions.
Why trust is the deciding factor
Patients come to AI in health care with real reservations. In Pew's national survey, 60% of US adults said they would be uncomfortable if their own provider relied on AI to diagnose disease and recommend treatment[1].
- Would be uncomfortable 60%
Source: Pew Research Center, 2022[1]
Show the data
| Item | Value |
|---|---|
| Would be uncomfortable | 60% |
Privacy is the other worry. In a 2026 KFF poll, 77% of the public said they were concerned about the privacy of personal medical information given to AI tools[4]. Physicians share the ambivalence: about 40% told the AMA they are equally excited and concerned about AI, with patient privacy and the patient-physician relationship among their top concerns[5].
Disclosure is how a practice shows it understands those concerns. It signals that the AI is a tool the practice chose and controls, used for specific tasks, with people behind it. Hiding it signals the opposite.
Where the law already requires it
In California, a health facility, clinic, physician's office or group practice that uses generative AI to create written or verbal patient communications about patient clinical information must include a disclaimer that the communication was AI-generated and clear instructions for contacting a human health care provider. For audio, the disclaimer comes verbally at the start and end of the interaction. The requirement does not apply when a licensed or certified provider reads and reviews the communication, and administrative matters such as appointment scheduling and billing are outside the definition of clinical information[2].
In Texas, the Responsible Artificial Intelligence Governance Act took effect on January 1, 2026. When an AI system is used in relation to a health care service or treatment, the provider must disclose that to the patient or their representative no later than the date the service is first provided, except in an emergency, when it must be given as soon as reasonably possible. The disclosure must be clear and conspicuous, in plain language and free of dark patterns[3].
For outbound calls, federal rules apply too. Artificial or prerecorded voice messages must state the identity of the business clearly at the beginning and give a callback number during or after the message[6]. Ask your counsel which rules reach your practice; the safe operating habit is the same everywhere.
What good disclosure looks like in each channel
| Channel | Where to disclose | Suggested wording |
|---|---|---|
| Inbound phone | First sentence of the greeting | “You've reached the practice's automated assistant.” |
| Outbound call | Opening line, with practice name and callback number | “This is an automated call from [practice name]. To reach us, call [number].” |
| Text message | Start of the first message in a thread; signature on later ones | “[Practice]: automated message.” or “Sent by our automated assistant” |
| Website chat | Persistent label on the chat window plus first message | “Automated assistant. Staff available during office hours.” |
| AI-drafted portal reply | If not reviewed by a clinician, label at the top | “This message was drafted with AI. To reach your care team, call [number].” |
Notice the pattern: the disclosure arrives before the patient has shared anything, it is in the same language and tone as the rest of the conversation, and it comes with a route to a person. A disclosure buried in a website privacy policy does none of those things.
Principles for writing the disclosure
- Say it first. One short sentence at the very start, before any question.
- Use ordinary words. Automated assistant is clearer to most callers than conversational agent or virtual receptionist.
- Pair it with the exit. Tell patients how to reach a person in the same breath.
- Do not apologize or oversell. Avoid both I'm only a robot and I'm just like a real receptionist.
- Repeat it when it matters. If the AI hands over to a person or a person hands back to the AI, say so.
- Answer honestly when asked. If a caller asks whether they are talking to a person, the answer is always a clear no.
What disclosure does not do
- It does not make clinical advice acceptable. A disclosed AI that interprets symptoms is still out of bounds.
- It does not replace consent for call recording where state law requires it. Recording notices are a separate requirement.
- It does not cover marketing calls or texts, which need their own consent.
- It does not excuse a bad experience. A disclosed AI that traps callers in loops still damages trust.
- It does not end your obligations under HIPAA. Vendors that handle patient information still need a business associate agreement.
Rolling disclosure out across channels
Inventory every AI touchpoint
List each place AI speaks or writes to patients: phone greeting, overflow line, reminders, text replies, website chat, portal drafts.
Write one wording standard
Agree a short phrase, such as automated assistant, and use it everywhere so patients learn it.
Mark the review status
For AI-drafted clinical messages, record whether a licensed provider reviewed them, since that changes what California requires.
Train staff to back it up
Front desk staff should be able to explain in one sentence what the assistant does and how patients can avoid it if they prefer.
Audit quarterly
Call your own line, send your own chat messages and read a sample of texts to confirm the disclosure still appears after updates.
When a patient says they would rather not
Disclosure invites a response, and some patients will tell you they do not want to deal with an automated system. Treat that as a legitimate preference, not a problem to overcome. A practice that respects it earns more goodwill than one that argues the technology's merits.
- On the phone, the route to a person should work every time, without the AI asking the caller to explain why.
- For texts, offer a no-automated-replies option in the patient's communication preferences, separate from opting out of reminders altogether.
- For portal replies, honor a request that a clinician write personally, and flag the record so drafts are not used for that patient.
- Record the preference once and apply it across channels, so the patient does not have to repeat it.
Front desk script: “Of course. The assistant helps us answer the phone faster, but you can always ask for one of us. I've noted on your file that you'd prefer to speak with the team, and when you call, just say receptionist and you'll come straight through.”
Track how many patients make this request. If the number is small and stable, your disclosure and design are probably working. If it rises after a change, such as a new greeting or a vendor update, look at what changed. Patients who opt out are also a useful panel: ask a few of them what put them off, and you will often hear about a specific frustration, such as repeated questions or a slow voice, that is easy to fix for everyone.
Example: a fictional multi-state group
Example: a fictional primary care group with clinics in California and Texas uses AI for its phone greeting, appointment reminders and drafts of portal replies. Rather than tailoring each channel to each state, it adopts one rule: every AI touchpoint discloses in its first line and gives a route to a person. For portal replies about clinical questions, the group requires a clinician to review every draft before sending, and records the reviewer's name.
Example, continued: the group's quarterly audit finds that a vendor update removed the disclosure from the after-hours greeting. Because the group checks its own lines, the problem is fixed within a day. The compliance lead adds a line to the vendor contract requiring notice before any change to scripted disclosures.
Frequently asked questions
Do I have to tell patients they are talking to AI?
In some states and situations, yes: California requires disclaimers on AI-generated clinical communications not reviewed by a licensed provider, and Texas requires providers to disclose AI used in health care services. Disclosing everywhere is simpler and builds trust.
What is the best wording for an AI disclosure?
Short and plain, at the start, with a route to a person: for example, you've reached the practice's automated assistant; say receptionist at any time to reach the team.
If a clinician reviews an AI-drafted message, does it still need a disclaimer?
Under California's rule, communications read and reviewed by a licensed or certified provider are exempt from the disclaimer requirement. Other states differ, so many practices label AI drafts regardless.
References
Pew Research Center (2022). 60% of Americans Would Be Uncomfortable With Provider Relying on AI in Their Own Health Care (opens in a new tab) ↑
California Legislature (code text via California Public Law) (2024). California Health and Safety Code § 1339.75 (added by AB 3030) (opens in a new tab) ↑
Texas Legislature (2025). H.B. 149, Texas Responsible Artificial Intelligence Governance Act (enrolled text) (opens in a new tab) ↑
KFF (2026). KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice (fielded Feb 24 - Mar 2, 2026) (opens in a new tab) ↑
American Medical Association (2026). More than 80% of physicians use AI professionally: AMA survey (opens in a new tab) ↑
Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 47 CFR § 64.1200 - Delivery restrictions (opens in a new tab) ↑