Almost everything written about AI on the phone assumes the machine is on the outbound side: your agent, your dialler, your list. The flip side is already live and barely discussed. Consumer-side assistants now ring businesses on a person’s behalf to check a price, confirm stock or book a slot. The call arrives on your main line, in business hours, sounding like a polite customer. It is not one.
Your front line was designed for humans. Hold music, four-level menus, “our team will call you back” — all of it assumes patience and memory. A consumer’s agent has neither. It has a task, a timeout, and a list of your competitors it is working through in parallel.
Quick answer: Consumer AI callers are real and confirmed: Google Search and Maps place automated calls to businesses to check pricing, availability, restaurant wait times and appointments, and Google publishes the disclosure opener its automated calls use — “This is an automated service from Google” — though the examples it prints are for the calls Google makes to collect business information, not for the calls it places on a customer’s behalf. Businesses should shorten IVR paths, state hours and prices as plain speech, publish structured data so the call never needs to happen, and log every caller that self-identifies as an agent.
Zian AI builds autonomous voice agents for the outbound side of this equation, so we have an interest in being honest about the inbound side rather than overselling it. The same constraints apply both ways: disclose early, resolve fast, hand off cleanly. To work through what should answer your phone, Apply For Partnership.
What is confirmed, and what is not
Confirmed at a Google primary. Google’s Business Profile help page, About automated calls and texts from Google to your business, states that when customers use Search or Maps to connect with a business, “Google might contact your business on their behalf through an automated phone call to complete the request.” The documented request types are appointment bookings, restaurant wait times, and price and availability. The calls “are monitored and recorded for quality assurance”, and in the US the feature runs in every state except Indiana, Louisiana, Minnesota, Montana and Nebraska. Google publishes no reason for the exclusions, so we will not invent one.
Disclosure is documented, with a caveat. Google’s page Help protect against fraudulent calls and texts prints actual openers, including: “This is an automated service from Google, and this call is recorded to help improve Google services.” Read the scope carefully: the examples Google publishes are calls it makes to collect business information — updating hours, checking restaurant attributes — and Google does not publish an equivalent script for the calls placed on a customer’s behalf. The same page says the calls come “from a set of designated numbers”, are “only for businesses that have public phone numbers for the purpose of receiving calls from the general public”, and can be stopped “by simply saying so on the call”.
Availability, precisely. The Search help page Use AI Mode to check local availability and pricing says the feature is “available in English to signed-in users over 18”, that Google “can make calls on your behalf” when a business has limited information online, and that call results “can take up to 30 minutes”. Its dining-reservations variant is listed, in English and Hindi, for the US, Australia, Canada, New Zealand, the UK, India, Singapore, Hong Kong, South Africa and four smaller jurisdictions (Anguilla, the Cook Islands, Gibraltar and the Isle of Man). The separate pricing help page confines quote-gathering to the US, and is blunt: “Using this service will not create a booking.”
What we could not confirm. As at 18 August 2026 we found no first-party documentation from Apple, OpenAI or Amazon describing a consumer feature that telephones a business on a user’s behalf. OpenAI documents plumbing, not product: its Realtime API with SIP guide says “With SIP and the Realtime API you can direct incoming phone calls to the API”, which is the receiving side. Treat “everyone will ship a calling assistant” as a forecast, not a fact. What is a fact is that Google already does it at consumer scale.
Human caller versus consumer AI caller
| Dimension | Human caller | Consumer AI caller | What your line should do |
|---|---|---|---|
| Hold music | Tolerates a minute or two, especially with a queue position | No social cost to abandoning; it is calling competitors in parallel | Answer common questions before the queue, not after it |
| IVR menus | Guesses, presses 0, or waits for a human | Parses the spoken options; wording that changes between calls breaks its plan | Stable wording, spoken keywords as well as DTMF, two levels maximum |
| Open-ended questions | Rephrases, negotiates, accepts “it depends” | Has one slot to fill; “it depends” returns as no answer | Give a number, a range with conditions, or an explicit “we quote after inspection” |
| Voicemail | Leaves a message and waits | Cannot usually act on a callback inside its task window; you drop out of the comparison | Put hours, price band and a booking URL in the greeting itself |
| Cost of a bad interaction | One lost lead, occasionally a review | Silent omission from a shortlist the customer actually reads | Instrument it: log agent-identified calls and unresolved outcomes as a named metric |
| Identity | Gives a name when asked | Announces the platform, and may pass the end customer’s name | Capture platform, end-customer name and callback channel as separate fields |
When their agent calls your agent
If you already run an AI receptionist on the front line, the AI-to-AI call is the default case, not a curiosity. Four things go wrong.
Double disclosure. Both sides announce themselves, then both wait politely. Google’s opener is scripted and consumes the opening of the call; yours does too. Keep your greeting short, front-loaded, and workable without a reply. Our guide to announcing an AI agent on a live call covers wording; the inbound rule is that the greeting must carry a fact, not just an apology for being a machine.
Loop and echo risk. Turn-taking, barge-in and silence thresholds are tuned to human speech rhythms. Two agents can settle into a courteous stalemate, each waiting out the other’s end-of-turn silence, or a clarification loop where neither may change the question. Set hard limits: maximum duration, maximum turn count, and a rule that three turns without new information ends the call with a spoken URL.
Authority verification. “Are you authorised to book on their behalf?” has a variable answer: Google’s pricing flow explicitly does not create a booking, whereas the Business Profile flow does book appointments once the customer has supplied details such as preferred time, service type and party size. Make policy per-action, not per-caller. Answer enquiries freely, issue quotes with an expiry, and confirm anything binding directly to the named end customer by SMS or email rather than to the agent. Never take payment details from an intermediary.
The structured-data escape hatch. The best evidence that a clean handoff beats a longer conversation comes from Google itself: if you have shared in-store product details via Merchant Center, “Google will use that information and will not call your business to ask about inventory.” The call is a fallback for missing structured data. Publish the data and the call does not happen — the phone-side mirror of our companion post on preparing your site and funnel for agents that browse and fill forms. When a conversation is unavoidable, end it with a bookable link, and make sure the handoff carries the context.
Does an AI caller have to say it is an AI?
Mostly no law forces it today — the FCC has proposed rules that would, but they are not in force — which makes Google’s voluntary disclosure notable rather than routine.
In the US, the FCC’s February 2024 Declaratory Ruling “recognizes calls made with AI-generated voices are ‘artificial’ under the Telephone Consumer Protection Act (TCPA)”. That sounds decisive until you read what the restriction attaches to. Under 47 CFR 64.1200(a)(1), the artificial-voice prohibition covers emergency lines, hospital and care-facility rooms, and “any telephone number assigned to a paging service, cellular telephone service, specialized mobile radio service, or other radio common carrier service, or any service for which the called party is charged for the call”. An ordinary business landline is not on that list; (a)(3) covers residential lines, and the telemarketing rules turn on the call carrying an advertisement, which an enquiry about your hours does not. If your published number is a mobile, the analysis is less comfortable, because (a)(1)(iii) reaches wireless numbers. One duty does survive the line-type question: 64.1200(b)(1) requires all artificial or prerecorded voice messages to “state clearly the identity of the business, individual, or other entity that is responsible for initiating the call” — an identification duty, not an admission of being a machine. This is our reading of which paragraphs bite, not a regulator’s ruling on agent enquiry calls, so take advice before relying on it.
The EU is where a duty bites. Article 50(1) of the AI Act, Regulation (EU) 2024/1689, requires providers to “ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system”, unless that is obvious. Your receptionist is a natural person, so an AI caller reaching an EU business is squarely in scope, and Article 113 sets the general application date at 2 August 2026 — now. The honest position: assume disclosure happens because the platform chooses it, not because it must, and build a line that copes when a future one does not.
The inbound readiness checklist
- Front-load the five machine-askable facts — hours, address, price band, whether you serve the category, next availability. Anything four menu levels deep does not exist. Our comparison of AI voice agents against IVR phone trees has the depth-versus-abandonment argument in full.
- Say prices and hours as plain speech. “Prices vary, speak to our team” is not an answer. A range with a stated condition is parseable; a deflection is not.
- Keep IVR wording stable and shallow. Announce the full option list once, accept spoken keywords as well as keypresses, and never bury the human beyond two levels.
- Publish the data upstream — Business Profile hours and services, Merchant Center inventory. Google’s documentation confirms this suppresses the inventory call entirely.
- Capture a callback identity. Google says it shares the requester’s details, “which may include your name”, with businesses it calls. The agent is not your lead; the person behind it is.
- Log the self-identification. Add a field: did the caller state it was an automated agent, and which platform did it name? Without that flag you cannot tell whether agent traffic is rising or whether you are losing it.
- Rehearse the human script. A receptionist who hangs up on an automated caller — whether it opens with “this is an automated service from Google” or simply asks about price and availability — may have removed you from a shortlist a real customer requested.
- Choose your opt-out deliberately. You can opt out on the call, in Business Profile advanced settings, or from Google’s phone-tree mapping calls via the voicemail line it publishes. That is a decision to be absent from a comparison, not a neutral act.
None of this requires an AI agent on your side, but it gets easier with one: software can be instructed to detect an agent caller, answer in structured facts, and close with a link instead of a promise. That is how Zian’s Customer Support and Appointment Setter agents are designed to behave. To get your line ready before this traffic compounds, Apply For Partnership.
Sources and who owns each figure
| Figure/claim | Owner (organisation) | Where it’s published | Date checked |
|---|---|---|---|
| Google places automated calls on customers’ behalf (bookings, wait times, price/availability); monitored and recorded; US-wide except IN, LA, MN, MT, NE; Merchant Center data suppresses inventory calls; opt-out routes including phone-tree mapping | Google (Business Profile Help) | support.google.com/business/answer/7690269 | 18 Aug 2026 |
| Disclosure openers (“This is an automated service from Google”); designated calling numbers; public business numbers only; opt out on the call | Google (Business Profile Help) | support.google.com/business/answer/6212928 | 18 Aug 2026 |
| AI Mode calls businesses when info is not online; English, signed-in, over-18; dining reservations in Australia, US, UK, NZ, Canada and others; results up to 30 minutes | Google (Search Help) | support.google.com/websearch/answer/17104441 | 18 Aug 2026 |
| US-only pricing flow; Google shares the requester’s name with businesses called; “Using this service will not create a booking” | Google (Search Help) | support.google.com/websearch/answer/16421135 | 18 Aug 2026 |
| AI-generated voices are “artificial” under the TCPA (Declaratory Ruling, 8 February 2024) | US Federal Communications Commission | docs.fcc.gov/public/attachments/DOC-400393A1.txt | 18 Aug 2026 |
| Line types the artificial/prerecorded-voice restriction applies to (47 CFR 64.1200(a)(1)–(a)(3)) | US Office of the Federal Register (eCFR) | ecfr.gov/current/title-47/section-64.1200 | 18 Aug 2026 |
| EU AI Act Article 50(1) disclosure duty; Article 113 application date 2 August 2026 | European Union (Regulation (EU) 2024/1689, EUR-Lex) | eur-lex.europa.eu — Regulation (EU) 2024/1689 | 18 Aug 2026 |
| “With SIP and the Realtime API you can direct incoming phone calls to the API” | OpenAI (platform documentation) | developers.openai.com/api/docs/guides/realtime-sip | 18 Aug 2026 |
Frequently asked questions
Is Google really calling businesses with an automated voice?
Yes, and Google documents it. Google Business Profile Help states that when customers use Search or Maps to connect with a business, “Google might contact your business on their behalf through an automated phone call to complete the request”, covering appointment bookings, restaurant wait times, and price and availability confirmation. The same page notes the calls are monitored and recorded.
Does the AI caller identify itself as a machine?
Google’s business-information calls do. Its page Help protect against fraudulent calls and texts publishes the openers, including “This is an automated service from Google, and this call is recorded to help improve Google services.” Google does not publish a script for the calls it places on a customer’s behalf, so do not assume every agent caller announces itself. No general US law compels an AI disclosure for an enquiry call to a business landline — 47 CFR 64.1200(b) requires an artificial-voice message to identify the business behind it, not to declare itself a machine — so treat platform disclosure as a choice. In the EU, Article 50(1) of the AI Act creates a genuine duty on providers of systems that interact directly with natural persons.
Is an AI call to my business a robocall under the TCPA?
Not usually. The FCC ruled in February 2024 that AI-generated voices count as “artificial” under the TCPA, but 47 CFR 64.1200(a)(1) attaches that restriction to emergency lines, patient and guest rooms, and wireless or charged-per-call numbers. An ordinary business landline is not on that list, and an enquiry about your hours is not telemarketing. If your published business number is a mobile, the position is less clear-cut.
Can I stop these calls?
Yes. Google’s Business Profile help page describes three routes: say so on the call (“Please stop calling my business”), toggle the settings under “Google automated calls and text messages” in Business Profile advanced settings, and separately opt out of Google’s calls to map your phone tree by leaving a voicemail on the number Google publishes. Weigh it up first — opting out removes you from comparisons a customer actively asked for.
Is this reaching Australia yet?
Partly. Google’s AI Mode help page lists the dining-reservations capability for the US, Australia, Canada, New Zealand, the United Kingdom, Singapore, India, Hong Kong, South Africa and four smaller jurisdictions, while the broader local pricing-and-availability calling flow is documented as US-only. Australian venues should expect reservation-driven agent calls before general enquiry calls.