The awkward question
Short answer: PhonePick tells them the moment they ask, on every plan, and there is no setting that turns that off — so the question mostly does not arise. The longer answer is worth reading, because a lot of what is written about this is wrong.
It is what we found when we read the actual legislation rather than other people's blog posts about it. Rules differ by state and province, several are new, and more are coming. If the answer matters to your business, ask someone qualified — and if you find something here that is wrong, tell us and we will fix it.
Search this question and you will find a dozen articles saying California requires you to disclose AI on calls, usually citing AB 2905. We read the bill. What it actually says is that a caller using an automatic dialing-announcing device must inform the person called if the prerecorded message uses an artificial voice.
That is a rule about outbound robocalls — machines dialling out and playing a recording at people. An AI receptionist answering a call that a customer chose to place to your business is a different thing, and AB 2905 does not appear to reach it.
Source: California AB 2905, approved 20 September 2024 — the bill text on leginfo.legislature.ca.gov. Read it yourself; it is short.
This matters because most of the alarm written about AI receptionists borrows its authority from robocall law. The two are not the same, and being told you are breaking a law you are not breaking is its own kind of harm.
There are real rules, and they are narrower and more specific than the blogs suggest. Utah requires a consumer-facing bot to say it is generative AI when asked, and requires proactive disclosure for regulated occupations — licensed professionals and certified services — without waiting for the question. Maine and the EU also have disclosure obligations. There is no single federal rule in the US, though the FCC has proposed one.
The picture is moving quickly enough that any table of states published today is out of date within months, which is exactly why we have not published one. If you want certainty for your state, that is a question for a lawyer and not for a software company's website.
It answers honestly the moment anyone asks whether they are talking to a machine. That is not a setting. There is no toggle for it, no plan on which it behaves differently, and no way for a customer to configure it away.
It also says so before anyone asks. Every call it answers opens by saying it is an AI and that the call is recorded, on every plan and for every kind of business, and that cannot be switched off either.
The reason is not compliance. It is that a receptionist which claims to be a person is one bad phone call away from being the story your customer tells about you — and the rules are only going to move in one direction from here.
A business with outbound calls can have PhonePick ring people for it: returning a missed call, reminding someone about an appointment, and on the Growth plan calling back a website lead or someone who agreed to hear from it. Here the robocall rules are the right ones to read, because this time the machine is the one dialling.
So every call PhonePick places announces itself first: the first thing the person hears is the assistant's name, that it is an AI assistant, and which business it is calling for — and on a sales call, straight after, how to opt out: “Hi, this is Nina, an AI assistant calling from Beckett Heating. If you’d rather not get these calls, just say “stop calling” at any time.” On calls it places there is no setting that turns that off, and it still answers honestly if asked again later in the call. A sales call never leaves a voicemail.
The FCC ruled in February 2024 that the TCPA’s rules on an “artificial or prerecorded voice” cover AI technologies that generate human voices, so a call made with one needs the called person’s prior consent, and a telemarketing call needs it in writing. In Canada, the CRTC’s Unsolicited Telecommunications Rules require express consent for sales calls that use a synthesized voice, as we read them. That is why PhonePick makes sales calls only to people with a consent record on file — when they agreed, the words they agreed to, and where — and why anyone who says not to call again is taken off there and then. The consent itself is the business’s to hold; what PhonePick enforces is set out in our terms.
Source: FCC Declaratory Ruling FCC 24-17, CG Docket No. 23-362, adopted 2 February 2024 and released 8 February 2024 — the ruling on docs.fcc.gov.
It never invents a price. It never promises something your calendar cannot honour. It says it does not know rather than guessing. Those are not legal requirements anywhere we have found — they are just the difference between a receptionist and a liability.
Two places outside a phone call, and the same honesty applies to both.
The chat button on a business's own website is answered by an AI: the same assistant as the phone, working from the same hours, services, prices and answers the business gave it. The chat window says it is an AI at the top and again at the bottom, and the assistant says so if anyone asks. It cannot book. When someone wants a time, a quote or a person, it takes their name and number and passes the message on; the business sees it in its leads and by email. It answers only from the business's own facts, and where those say nothing it says it isn't sure and offers to take a message rather than guess. It won't take a card number.
When a business texts a photo of a receipt to its PhonePick number, an AI reads it. The tax it records is the tax printed on the receipt, never a figure worked out from a rate. Where no tax line is printed but a subtotal is, the difference is used and marked for a person to check; with neither, no tax is recorded. Only photos from the owner's phone and the staff phones the business lists are read this way, so a customer's photo stays an ordinary message.
PhonePick