This is a plain-language overview, not counsel. AI-disclosure law is young and varies by state and context. For your specific situation, check with a lawyer who knows your jurisdiction.
Why this is suddenly a question
Voice AI got good enough, fast enough, that a caller often can’t tell they’re not talking to a person. Regulators noticed. A handful of states have moved to require that automated or AI-driven callers disclose what they are in certain contexts, and more proposals are in the pipeline. The direction of travel is clear even where the letter of the law isn’t yet.
Inbound is a gentler case than outbound
Most of the sharpest rules target outbound automated calls — robocalls, AI-generated voices dialing you. A restaurant phone agent is the opposite: it answers calls the customer chose to place. That’s a lower-risk posture. It does not make disclosure obligations disappear, but the customer initiating contact is a meaningfully different situation than a bot cold-calling them.
The case for saying it anyway
Set the law aside for a second. A brief, natural “you’ve reached our virtual assistant” is cheap, and where an owner chooses to use it, it tends to earn its keep:
- It stays ahead of disclosure rules instead of testing their edges.
- It sets the caller’s expectations, so the interaction goes smoother.
- It avoids the one genuinely bad outcome — a customer who feels tricked.
The counterweight is real too: some owners find an opening announcement makes a warm line feel procedural, and a regular calling for the fourth time this month does not need to be told again. That is a judgement about your own customers, which is why it is set per restaurant rather than decided for you.
Recording is a separate obligation
Don’t conflate two things. Whether the agent is AI is one question; whether the call is recorded is another, with its own — often stricter — consent rules that turn on which state the caller is in. We mapped those separately in two-party consent and restaurant call recording. If you record, handle that consent on its own terms.
Worth being concrete, because the two get blurred constantly: R.ai records every inbound call, and R.ai does not play a spoken recording announcement at the start of those calls. There is no dashboard setting to turn one on. Whether your line needs one is a separate question from the AI question. It turns on where your callers are dialing from, and it is yours to settle with your own lawyer.
Where R.ai actually lands
On the AI question specifically, the choice is the restaurant’s and we ask for it during onboarding: should the agent introduce itself as an AI on a caller’s first call? — with the box checked to start — and separately, should it do so for repeat callers. Turn it on and the agent is instructed to mention it naturally early in the call rather than reciting a script. Turn it off and the agent is instructed never to claim to be a person if a caller asks. What we don’t do is decide it for you, and we don’t play a fixed announcement over every line.
Because that instruction reaches the caller through a language model rather than a recorded clip, treat it as the agent’s standing brief, not a guaranteed sentence. If your jurisdiction requires specific, reliable wording on every call, say so before you go live — and check it with counsel rather than with us.