Key takeaways
- Since February 8, 2024, calls made with AI-generated voices count as artificial voice calls under the TCPA[1].
- The ruling is about calls your practice places. Outbound AI reminders and outreach need the same consent and identification as prerecorded calls.
- Telemarketing robocalls need prior express written consent[1]; certain healthcare calls to mobile numbers are exempt only if strict conditions are met[2].
- Damages are $500 per violation or actual loss, whichever is greater, and can be increased for willful violations[3].
What the FCC decided
On February 8, 2024, the Federal Communications Commission ruled unanimously that calls made with AI-generated voices are artificial under the Telephone Consumer Protection Act, effective immediately[1]. The TCPA has long restricted calls that use an artificial or prerecorded voice. The ruling closed any argument that a synthetic voice generated in real time, or a cloned voice, falls outside those restrictions.
For medical and dental practices, the ruling matters because AI voice tools are increasingly used for outbound work: appointment reminders, recall campaigns, confirmations, no-show follow-ups and waitlist offers. Those calls now sit firmly inside the same rules as the prerecorded reminder calls many practices have used for years. The good news is that those rules are well established. The risk is that new tools make it easy to place many calls quickly, so a mistake scales just as fast.
Inbound and outbound are different questions
The TCPA's restrictions on artificial voice calls apply to calls a business initiates. When a patient calls your practice and an AI voice answers, that is not a robocall under the ruling, although other rules still apply, including state call recording consent laws and state AI disclosure laws. When your system calls a patient, the TCPA applies in full.
That distinction drives a lot of practical design. Many practices start with inbound AI, such as overflow and after-hours answering, precisely because the consent questions are simpler. Outbound AI calling should be treated as its own project, with counsel review of the consent basis for each type of call before anything is switched on.
Consent by call purpose
| Call purpose | Typical consent basis | Notes for practices |
|---|---|---|
| Emergency purposes | Exception in the rules | Narrow; do not stretch it to routine outreach |
| Exempt healthcare message to a mobile number | No prior consent if every exemption condition is met | Appointment and lab result notices are typical; no billing or marketing content |
| Other informational calls to a mobile number | Prior express consent | Often given when the patient provides their number for contact; confirm with counsel |
| Telemarketing or advertising | Prior express written consent | Promoting a new service, cosmetic offers, membership plans |
| Billing and collections | Separate analysis | Excluded from the healthcare exemption's content; get advice |
The general rule in the FCC regulations is that no one may initiate a call using an artificial or prerecorded voice to a mobile number without prior express consent, other than for emergency purposes[2]. For telemarketing, the bar is higher: prior express written consent[1]. With about 98% of US adults owning a cellphone[4], assume nearly every patient number on file is a mobile number.
The healthcare exemption, applied to voice calls
FCC rules include an exemption for healthcare calls to mobile numbers that are not charged to the patient. A practice relying on it for AI voice calls should be able to show each of these conditions for every call[2]:
- The call goes only to the mobile number the patient gave the practice.
- The call states the practice's name and contact information.
- Its purpose is one the rule allows, such as an appointment reminder or a notice that lab results are ready.
- It contains no telemarketing, solicitation or advertising, and no accounting, billing, debt-collection or other financial content.
- The practice places no more than one such call or text per patient per day, and no more than three per week, combined across voice and text.
- The call is concise, generally one minute or less.
- Each call offers an easy way to opt out; for live-answered calls, an automated opt-out mechanism.
AI voice tools tend to break conditions four, five and six. A model that improvises can wander into a balance reminder or a promotion. Separate campaigns each sending one message a week can add up to more than three. And a conversational voice that answers follow-up questions can easily run past a minute. Configure hard limits rather than relying on the script.
Identification and opt-outs
Every artificial or prerecorded voice message must state clearly, at the beginning, the identity of the business placing the call, and give a callback telephone number during or after the message[2]. Patients may revoke consent by any reasonable method, including using the call's automated opt-out or replying stop to a text, and the request must be honored within a reasonable time not exceeding ten business days[2].
Reminder script for an AI voice: βHello, this is an automated call from [practice name] at [number]. This is a reminder of an appointment on [day] at [time]. Press 1 to confirm, press 2 to speak with our team, or press 9 to stop these calls.β Keep it under a minute, and say nothing about the type of visit.
Why the liability adds up
The TCPA lets people sue for their actual loss or $500 per violation, whichever is greater, and courts may increase damages for willful or knowing violations[3]. Because each call can be a separate violation, exposure grows with volume, which is exactly what automated calling produces.
Source: Legal Information Institute, Cornell Law School, current[3]
Show the data
| Item | Value |
|---|---|
| 50 calls | $25,000 |
| 250 calls | $125,000 |
| 1,000 calls | $500,000 |
The chart is an Example built only from the $500 statutory figure multiplied by fictional call counts. It ignores any increase for willful violations and legal costs. The point is scale: a misconfigured campaign that calls the wrong list for a week can create exposure far larger than the cost of a careful review before launch.
Why outbound AI calling is tempting, and what to put in the contract
The pull toward automated outreach is real. In an MGMA poll, 32% of medical groups said patient no-show rates were higher in 2026 than in 2025, while 58% said about the same and 10% lower[5]. Reminder and confirmation calls are a standard response, and AI voices promise calls that can handle a reschedule in the same conversation rather than just playing a message.
That extra capability is also where the TCPA risk grows, because a conversational call is harder to keep short, on-topic and free of financial content. If you buy outbound AI calling from a vendor, put the controls in the contract rather than relying on a sales call:
- The vendor configures per-patient frequency caps you set, and reports any call placed above them.
- Scripts and allowed topics are approved by the practice, and the voice cannot add billing, promotional or clinical content.
- Calls identify the practice and give a callback number at the start, every time.
- Opt-outs by keypress, voice or text are processed across all your campaigns, with a log you can export.
- Call logs include time, number, script version and outcome, and are kept for the period you specify.
- The vendor notifies you before changing the voice model or call flow, and you can test changes first.
- Responsibility for compliance failures caused by the vendor's configuration is addressed explicitly.
None of these terms replaces legal advice on your consent basis, but together they make sure the tool behaves the way your counsel has approved.
Auditing an outbound AI calling program
List every outbound call type
Reminders, confirmations, recall, no-show follow-up, waitlist offers, surveys and any marketing.
Record the consent basis
For each type, write down whether it relies on the healthcare exemption, prior express consent or written consent, and where that consent is stored.
Lock the content
Approve scripts and prevent the AI voice from adding billing, promotional or clinical content.
Set frequency caps across systems
Count voice and text together per patient per day and per week, across every vendor.
Test opt-outs end to end
Opt out on a test number by keypress, by voice and by text, and confirm all systems stop within your internal target.
Keep records
Log each call, its script version, the consent basis and any opt-out, and keep the logs for the period counsel advises.
Example: a fictional recall campaign
Example: a fictional dental group plans to use an AI voice to call patients overdue for cleanings, with an offer of a discounted whitening add-on. Its compliance review flags the problem: the whitening offer is advertising, so the call is telemarketing and needs prior express written consent, which most patients have not given. The group splits the campaign: an informational recall reminder within the healthcare rules, with no offer, and a separate written-consent program for promotions.
Example, continued: in testing, the AI voice answers a patient's question about a past balance. The group removes billing topics from the outbound voice entirely and routes any such question to a staff callback. It also discovers that its reminder and recall vendors were counting messages separately, so one patient could receive four contacts in a week. A shared cap fixes it. Patients' general views on automated voices are covered in will patients talk to an AI voice.
Frequently asked questions
Are AI voice calls considered robocalls?
Yes, for TCPA purposes. The FCC ruled on February 8, 2024 that calls made with AI-generated voices are artificial voice calls, so the same consent, identification and opt-out rules apply as for prerecorded calls.
Can a medical practice use AI voices for appointment reminders?
Yes, if the calls meet the TCPA rules: either a valid consent basis or every condition of the healthcare exemption, including frequency caps, a one-minute limit, no marketing or billing content, identification and an easy opt-out.
Does the FCC ruling apply when patients call us and an AI answers?
The ruling concerns calls a business places. Inbound AI answering is not a robocall, though state recording consent and AI disclosure laws can still apply.
References
Federal Communications Commission (2024). FCC Makes AI-Generated Voices in Robocalls Illegal (news release) (opens in a new tab) β
Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 47 CFR Β§ 64.1200 - Delivery restrictions (opens in a new tab) β
Legal Information Institute, Cornell Law School (current). 47 U.S. Code Β§ 227 - Restrictions on use of telephone equipment (opens in a new tab) β
Pew Research Center (2026). Americans' Use of Mobile Devices in 2026 (Mobile Fact Sheet) (opens in a new tab) β
Medical Group Management Association (MGMA) (2026). About 1 in 3 medical groups see higher no-shows in 2026 as patients face higher costs (opens in a new tab) β