An office manager at a two-dentist practice near Tampa has a vendor contract open and one question her owner keeps asking: are we even allowed to do this? The draft greeting says “Hi, this is Ava from the front desk.” The vendor records every call. And the plan includes calling back anyone who hangs up.
Each of those three details touches a different rule. None makes AI phone answering illegal, but each needs a decision. Here they are in the order a caller would meet them.
Flowchart
Four checks before the assistant takes its first call
- StartDraft greeting and vendor contract on the table
- DecisionDoes the first sentence say the assistant is automated?NoStepRewrite it, and drop human names and “front desk” phrasingYes
- DecisionAre calls recorded or transcribed?NoStepGet “no audio or transcript kept” in writing; no recording notice neededYes
- StepRecorded: announce it in the greeting, on every call, in every state
- DecisionWill the assistant ever place calls or send texts?YesStepSettle consent, caller identification and opt-outs firstNo
- DecisionWill callers share health or legal details?NoStepStandard confidentiality terms in the vendor contractYes
- A personYour lawyer or privacy officer reviews the BAA or confidentiality terms
- DoneLaunch, then review a sample of calls every month
1. Tell callers they're talking to an automated assistant
No single federal law requires an inbound phone assistant to announce that it's AI. Some states now require disclosure in specific situations:
- Maine bars using an AI chatbot or other computer technology in trade in a way that could mislead a reasonable consumer into believing they're dealing with a human, unless the consumer is clearly and conspicuously told otherwise. The definition covers spoken conversation as well as text, and a violation counts as an unfair trade practice.
- Utah requires a business using generative AI with consumers to disclose it when the person clearly asks whether they're talking to a human. Licensed professionals in state-regulated occupations must disclose it up front, verbally at the start of a call, when the use is a “high-risk” interaction under the law.
- California requires clinics, health facilities, physician offices and group practices that use generative AI for patient communications about clinical information to say so, spoken at the start and end of an audio interaction, and explain how to reach a human, except where a licensed provider has reviewed the message. Scheduling and billing fall outside the rule.
The practical answer is simpler than the patchwork: disclose everywhere. A first sentence like “I'm the practice's automated assistant” covers most of what these laws ask, and it avoids the bigger business risk of a caller feeling tricked. That's why “Ava from the front desk” is the first line to change.
2. Recording and transcripts need the right consent
Most AI receptionists record or transcribe calls, because that's how you get a written summary. Federal wiretap law allows recording when one party to the call consents, and many states follow that one-party approach. Others require everyone's consent.
California prohibits recording a confidential communication without the consent of all parties. Florida's law allows interception when all parties have given prior consent. Maryland's statute likewise requires all parties' prior consent. Washington requires the consent of all participants, but treats consent as obtained once callers have been told, in a reasonably effective way, that the call is being recorded, and the announcement itself has to be captured on the recording.
Callers don't always dial from your state, so the safest habit is to announce recording at the start of every call, whatever your own state's rule. Because a vendor is listening on your behalf, it also helps to say so: “This call is recorded and transcribed by our service provider.” Then read the contract for what happens to recordings afterward: who can access them, how long they're kept, and a clear bar on the vendor using your call data for its own purposes, including training models.
3. Calling or texting people back is a different rulebook
Answering a call someone placed to you is one thing. Placing a call with an AI voice is another. A February 2024 FCC ruling held that the Telephone Consumer Protection Act's limits on “artificial” voices cover voices generated by AI, so calls using them face the same consent rules as other robocalls. For telemarketing calls, that means prior express written consent.
FCC rules also require artificial or prerecorded voice messages to identify the business at the start and give a callback number. So if the plan includes calling back every hang-up, treat it as an outbound program: who agreed to it, what the message says, how people opt out. Keep automated callbacks to callers who asked for one, note that request in the summary, and get your lawyer's view on anything broader. Texts raise similar questions; read the TCPA texting rules before switching them on.
4. Health and legal information need contracts, not promises
For dental and medical practices, the vendor that processes your patients' calls becomes a HIPAA business associate as soon as it handles their health information for you. Get a signed Business Associate Agreement before the first live call. A vendor that won't sign one isn't an option, however good the demo.
Law firms carry a parallel duty. ABA Formal Opinion 512 says the familiar duties (competence, confidentiality, client communication, supervision) follow lawyers into their use of generative AI, which in practice means knowing where call data goes and supervising what the assistant says. Intake calls also raise a conflict-of-interest issue, covered in our post on law firm intake after 5 PM.
5. You're answerable for what it says
If the assistant tells a caller something wrong, “the AI said it” won't get you far. In the 2024 case Moffatt v. Air Canada, a Canadian tribunal held the airline responsible for inaccurate bereavement-fare information its website chatbot gave a customer, and rejected the airline's argument that the bot was a separate entity answerable for itself. It's a Canadian small-claims decision, not binding in the US, but the reasoning is easy to imagine anywhere.
Keep the assistant's answers to information you've approved, and have it say “I'll have someone confirm that” for everything else.
Before and after
A risky opening vs. a careful one
Risky
- 01“Hi, this is Ava at the front desk!” sounds like a person
- 02No mention that the call is recorded
- 03Guesses at “Will insurance cover my crown?”
- 04Calls back every hang-up with an AI voice
ResultDisclosure, recording and consent problems on a single call
Careful
- 01“You've reached [Practice]'s automated assistant.”
- 02“This call is recorded and transcribed by our service provider.”
- 03Coverage questions go to staff, with a callback time
- 04Automated callbacks only for callers who asked
ResultThe caller knows who they're talking to and what happens next
A greeting and launch checklist
Adapt this opening: “Thanks for calling [Business]. I'm [Business]'s automated assistant, and this call is recorded and transcribed by our service provider. I can book appointments and answer common questions, and if you'd rather speak with someone, just say so.”
- The first sentence says “automated assistant,” and the recording notice comes before any questions.
- Saying “person” or “representative” reaches a human, or a message for one, every time.
- The contract covers recording access, retention, and a bar on the vendor using your call data for its own purposes, including training models.
- A Business Associate Agreement is signed (health care), or confidentiality terms are reviewed (law firms).
- Outbound calls and texts stay off until consent and opt-out handling are settled.
- The assistant answers only from information you've approved.
- A lawyer licensed in your state has reviewed the setup.
If you'd like a phone assistant built to this checklist, with your lawyer signing off on the state-specific parts, that's where we start when we build an AI receptionist.
Sources
- FCC: FCC Makes AI-Generated Voices in Robocalls Illegal (Feb. 8, 2024)
- FCC: Declaratory Ruling FCC 24-17, CG Docket No. 23-362
- 47 CFR 64.1200(b): identification requirements for artificial or prerecorded voice messages
- 18 U.S.C. § 2511(2)(d): one-party consent under federal law
- California Penal Code § 632
- Florida Statutes § 934.03
- Revised Code of Washington 9.73.030
- Maryland Courts and Judicial Proceedings § 10-402
- Maine Revised Statutes, Title 10, § 1500-DD
- Utah Code § 13-77-103: required disclosures for generative AI
- California Health and Safety Code § 1339.75 (AB 3030)
- HHS: Business Associates (HIPAA guidance)
- ABA Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024)
- Moffatt v. Air Canada, 2024 BCCRT 149