Does federal law make an AI receptionist say it is AI?
Not yet. In February 2024 the FCC ruled that an AI-generated voice counts as an "artificial or prerecorded voice" under the Telephone Consumer Protection Act5. That ruling is about calls a business places. A call you make to a customer in an AI voice generally needs that person's prior consent, just like any other robocall5. It says nothing about software that picks up when a customer calls you.
In August 2024 the FCC proposed going further. Anyone calling with an AI-generated voice would have to say so at the start of the call, and say it again when asking for consent1. The proposal is written for outbound calls only. In its own words, the TCPA's requirements "do not extend to technologies used to answer inbound calls"1. It is still a proposal: we found no final rule in the FCC's docket when we checked on 3 November 2026.
Which states require a business to say it is using AI?
Maine has the clearest rule. Its 2025 law bars using an AI chatbot with customers "in a manner that may mislead or deceive a reasonable consumer" into thinking they are dealing with a human, unless the customer is told clearly and conspicuously that they are not2. The law defines a chatbot as a program that imitates conversation through "textual or aural communications", so a voice agent on your phone line fits the words2. A breach counts as a breach of Maine's Unfair Trade Practices Act2.
California's bot law is narrower. It makes it unlawful to use a bot to talk with someone in California online while hiding that it is a bot, in order to push a sale or sway a vote, and a clear disclosure is a complete defense6. The law defines online as a public website or app6, so it fits website chat more naturally than a phone call.
Colorado changed course. Its 2024 AI Act said a business deploying an AI system that talks with consumers must tell each one they are dealing with AI3. In May 2026 the legislature repealed and rewrote that part of the law with SB 26-189, which takes effect on 1 January 20274. The new text covers automated tools that shape big decisions about jobs, housing, loans, insurance or health care, and it says routine scheduling and customer service triage are not such decisions4.
Utah also has a law on telling consumers about generative AI. The Utah Legislature's website would not load for us on 3 November 2026, so we have not summarized it here. Read it on the legislature's own site before you rely on anyone's summary of it, including the vendor blog posts that rank for this question.
| Rule | What it covers | What it asks for | Status on 3 November 2026 |
|---|---|---|---|
| FCC ruling 24-17 | Calls a business places in an AI voice | Prior consent, as for any artificial or prerecorded voice call | Adopted 2 February 2024 |
| FCC proposal 24-84 | Outbound calls and texts that use AI content | Say the call uses AI at the start, and when asking for consent | Proposed August 2024; no final rule found |
| Maine Public Law 2025, chapter 294 | AI chats, typed or spoken, used in trade with consumers | Clear notice that the customer is not talking to a human, where the chat could mislead | Signed 12 June 2025 |
| California Business and Professions Code 17941 | Bots online used to sell or sway a vote | Clear disclosure that it is a bot | In the code today |
| Colorado SB 24-205 | AI systems that interact with consumers | Tell each consumer they are dealing with AI | Repealed and rewritten by SB 26-189 |
| Colorado SB 26-189 | Automated tools that shape big decisions such as jobs, housing or loans | Notice before use, and an explanation after a bad outcome | Takes effect 1 January 2027 |
Is it enough to answer honestly when a caller asks?
Answering honestly when asked is the minimum. Maine's test is whether the conversation may mislead a reasonable customer2, and that customer may never think to ask. A warm, natural voice that greets people like a staff member and admits it is software only when pressed is exactly the case that test is aimed at.
Saying it once, up front, removes the question for every caller. It costs a few seconds. If your line already plays a recording notice, put the AI line in the same sentence, so the caller hears one short opening instead of two separate announcements before they speak.
What should the AI say when it answers?
Keep it short and put it first. A workable opening names the business, says what is answering and mentions the recording: "Thanks for calling Rivera Plumbing. You've reached our virtual assistant, and this call is recorded. How can I help?" Use plain words such as virtual assistant or automated assistant. Avoid a human first name that suggests a staff member, unless the next words make clear it is software.
Then test it yourself. Call your own number from a phone that is not saved as a contact, listen to the first ten seconds, and ask "Am I talking to a real person?" partway through. The answer should be a plain no. If it hedges, fix the script. Repeat the test after every change to the greeting, because a rewritten script can quietly drop the line.
These rules keep changing: the FCC proposal is still open, and Colorado's new law starts in 2027. This page is general information, not legal advice. If you work in a licensed field or take calls from customers in many states, ask a lawyer licensed where you operate.