What does US federal law actually require today?
Federal law regulates outbound calling far more tightly than inbound answering. In February 2024 the FCC confirmed that AI-generated voices in robocalls count as "artificial" under the Telephone Consumer Protection Act, which means calls you place using a synthetic voice generally need prior express consent. That ruling is about calls a business makes, not calls it receives. The Commission has an open rulemaking on AI-generated voice disclosure that would extend transparency duties further, but until it concludes there is no federal statute compelling you to announce AI to someone who dialled your number. Treat the direction of travel as settled and the timing as unresolved.
Which state laws are commonly cited, and do they apply?
Two California statutes come up constantly and neither says what marketing pages claim. SB 1001, the Bolstering Online Transparency Act in force since July 2019, prohibits using an undisclosed bot to incentivize a sale or influence a vote — but it is written for online communications and applies to public-facing platforms with at least ten million monthly US visitors, so it does not by its terms reach a local business's phone line. SB 243, effective January 2026, imposes real disclosure and safety duties but is scoped to companion chatbots, meaning systems designed to sustain social relationships, not a receptionist that books appointments. Utah's disclosure law is the closest fit: it requires a business to confirm it is AI when a consumer asks.
- CA SB 1001 — online bots, large platforms; not a phone-line rule
- CA SB 243 — companion chatbots; a receptionist is out of scope
- Utah — must answer truthfully when the consumer asks
- EU AI Act Art. 50 — applies from 2 August 2026 for EU-facing systems
Why does call recording change the answer entirely?
This is the part that decides it in practice. Almost every AI phone system records and transcribes calls, because that is how it works. Twelve states require the consent of all parties before a call may be recorded, and in California the requirement can attach when either party is in-state. Consent in those states means the caller is told, at the start, that the call is recorded. Once you are already announcing a recording, withholding the fact that the voice is synthetic is a strange line to draw and an easy one to attack later. The operationally simple posture — and the one most vendors adopt — is a single disclosure that covers recording and AI together, played on every call nationwide.
What should the disclosure actually say?
Short, first, and unambiguous. A sentence at the top of the greeting that names the business, states that the assistant is an AI, and states that the call may be recorded satisfies both duties without turning the greeting into a legal notice. Three failure modes to avoid: burying it after the caller has already started explaining their problem, making it removable by the business owner in settings, and phrasing it so vaguely that a reasonable caller would still believe they reached a person. If a caller asks directly whether they are talking to a robot, the system must answer honestly — that is the one requirement that is unambiguous almost everywhere.