When a customer's AI agent calls your business, treat it as a messenger. Tell it what you would tell any stranger: your hours, your services, what is available. Take the booking as provisional and confirm it with the customer in writing. Never hand it a customer's details, take card numbers from it, or agree to anything you could not undo. And do not hang up.
Most owners expect their first AI agent to be something they buy. For many it will be something that rings the front desk on a Tuesday, says it is calling for a Mrs. Alvarez, and asks whether you can fit in a brake inspection on Friday. Whoever picks up has about four seconds to decide what to do, and nobody has told them.
Sources checked November 16, 2026.
Are AI agents really calling small businesses?
Yes, in the United States, and the callers come from more than one company.
Google's help page for business owners says that when customers use Search or Maps, "Google might contact your business on their behalf through an automated phone call to complete the request." It lists three jobs: booking appointments, checking restaurant wait times, and confirming the price and availability of products and services (Google Business Profile Help, read November 16, 2026). Google announced the price-and-availability calls for all US Search users on July 16, 2025 (Google, July 2025). The same help page says the feature works only in select regions and languages, and in the US in every state except Indiana, Louisiana, Minnesota, Montana and Nebraska.
Two more arrived this fall. On September 17, 2026, TechCrunch reported that Meta's Muse assistant had gained the ability to place calls to US businesses, and that the startup Instinct had launched a calling feature for jobs such as booking restaurants that take no online reservations or getting on a dentist's cancellation list (TechCrunch, September 2026). Reuters described Muse users sending it to book haircuts, check whether a store has an item in stock, and collect quotes from contractors (Reuters via BNN Bloomberg, September 22, 2026).
Here is the part that should change how you think about it. A survey of 1,000 UK small business owners, run in August 2026 and released on September 17, 2026 by the Small Business Institute with Alibaba.com, found that 44% had heard of agentic AI and only 5% used it regularly (Small Business Institute, September 2026). Very few owners run an agent. Their customers do not need permission to send one. You may meet your first agent from the wrong side of the counter.
How do you know the caller is an AI agent?
Sometimes it tells you, and sometimes you cannot know, so your routine must work either way.
Google's own assistant is the clear case. Its help page for the Gemini app says: "Gemini starts every call by disclosing that it's an AI assistant from Google calling on a recorded line on your behalf, stating your name" (Gemini Apps Help, read November 16, 2026). The "your" there is the customer, so the call opens with the customer's name. The same page says Gemini cannot make payments or share card numbers, and cannot relay sensitive personal information such as passwords or health details.
The other callers are less tidy. Reuters reported, from internal company posts, that Meta had tested having human contractors quietly place some of the calls requested through Muse. A Meta vice president wrote that "it was a miss" to start that test without proper disclosures, and the company rolled it back. One employee in the same posts said of his insurer: "they keep hanging up on Muse when they hear it is AI."
Read those two details together. The voice asking about Friday could be software, or a contractor in a call center working for the software. Your staff cannot tell which from the sound, and they do not need to. A rule that depends on spotting the robot will fail. A rule that depends on what is being asked will hold.
What is the Messenger Rule?
The Messenger Rule is my suggested way of deciding, in the moment, what an agent may have. It is my way of thinking about the problem, not established research.
The Messenger Rule is how I suggest a small business treats any AI agent that contacts it for a customer: as a messenger. Give it what you would give a stranger who rang on that customer's behalf, which is public information and a provisional booking. Anything that needs the customer's identity, money or consent goes back to the customer, on contact details you hold for them.
A messenger can carry a request. A messenger cannot prove who sent them, cannot pay, and cannot agree to terms for someone else. You already know how to deal with one, because people have always phoned on behalf of a spouse, a boss or an elderly parent.
Here is the rule applied to the requests an agent is likely to make.
| The agent asks | What your staff do | Why |
|---|---|---|
| Hours, location, whether you do a certain job, what is available | Answer plainly | It is public. You would tell anyone. |
| A quote for a standard job | Give your standard figure and say what could change it | Google says the customer receives your quote by email or text, so it will be in writing |
| To book a new appointment | Book it as provisional. Take the customer's name and their own phone number or email. Confirm to the customer directly | Your written confirmation is the record of what was agreed |
| To change or cancel an existing booking | Tell the agent you will confirm with the customer, then do so on the details you already hold | Only the customer can give up their slot |
| Anything about an existing customer: their address, order, history or balance | Decline, and ask that the customer contact you | It is personal information, and a messenger has not proved who sent it |
| To pay, to read out card details, or to arrange a refund | Never on the call. Send your usual payment link to the customer | Google says its callers never ask for payment. One that does is a warning sign |
| A complaint, a negotiation, or anything odd | Take a message. A person calls the customer back | Judgment calls belong to people on both ends |
Take a hypothetical 12-person auto repair shop. The agent asks whether a brake inspection can be done on Friday and what it costs. The service adviser gives the standard figure, offers 10:30, asks for Mrs. Alvarez's own cell number, and says a text will follow. The text goes to her, she replies yes, and the slot is hers. If the agent had then asked what was done to her car last spring, the answer would be: "She is welcome to call us for that."
What should your staff say when one calls?
Your staff should say one sentence, then follow a card. The sentence is: "Thanks. Who are you calling for, and what have they asked you to do?"
That question does three jobs. It gets you a customer name for the booking. It tells you how far the request goes. And it works just as well when the caller is a person.
Then give the front desk a card. Copy this, fill it in, and tape it beside the phone.
The Front Desk Card
- What we help an agent with: ______ (for example: new bookings, hours, services, availability, standard quotes)
- What we only do with the customer: ______ (changes, cancellations, account details, anything with money)
- What we ask every time: who it is calling for, what it has been asked to do, and the customer's own number or email
- How we confirm: ______ (a text or email to the customer the same day, in our words)
- Who decides when something goes wrong: ______ (one name)
- What we write down: the date, what it asked for, and who it said it was calling for
The card is the Readiness Test at the size of a phone call. The Readiness Test is three questions an organization must be able to answer consistently, from the top to the front line, before it can call itself AI ready: What is AI for here? What may I do with it? Who decides when it goes wrong? Lines one and two answer the first two for this one situation. Line five answers the third. If you have not yet said those answers out loud to your team, a five-minute briefing is the place to add this.
One more thing for the card's margin: assume the call is being recorded. Google's page for owners says of its customer calls, "These calls are monitored and recorded for quality assurance," and the Gemini help page says the customer can read the transcript and listen to the recording afterward. Your customer may hear exactly how your front desk treated their assistant. If your own team records calls, the same courtesy applies in reverse, which I cover in the Open Room Rule.
Should you just block AI callers?
No, unless your staff cannot answer booking, stock or availability questions by phone at all. Here are the three approaches a small firm can take, compared on November 16, 2026. The table is my reading, not a survey finding.
| Approach | What happens | What it costs you |
|---|---|---|
| Hang up, or opt out everywhere | The agent reports back that you could not be reached, and tries the next business | The booking, and a customer who never learns why |
| Treat the agent as if it were the customer | Bookings are quick | Changes, account details and payments are agreed with something that has proved nothing |
| The Messenger Rule | Public questions are answered, new bookings are taken, the rest goes to the customer | One confirming text per booking |
My verdict: serve the messenger and confirm with the customer. The first approach turns away work that was already on its way to you. The second is how a slot gets given up, or a home address read out, on the say-so of a voice.
If you do want out, Google is the one caller with a published switch. Its help page says you can tell the caller "Please stop calling my business," or, for a verified Business Profile, open Business Profile settings, choose Advanced settings, and under "Google automated calls and text messages" turn off "Bookings and inquiries from customers." Google's announcement puts it this way: "businesses are always in control via their Business Profile settings." I found no equivalent published opt-out for Muse or Instinct in the reporting I read.
Before any of this matters, the agent has to find you. It works from your listing and your website, so the plain facts there decide whether you get the call at all. That is the subject of how to get your business recommended by AI.
What about scams that sound like agents?
Treat any caller who wants money or passwords as a scam, whatever it says it is. Google's page is blunt: "Google will never ask for payment, credit card information, or account passwords. If a caller asks for money to keep your business on Google, it is not a call from Google."
The Messenger Rule already covers this, because a messenger never gets money or account details. For requests that do involve payment, such as a "supplier" with new bank details, use the Known Number Rule in how to stop AI voice and payment scams: call back on a number you held before the request arrived.
What about agents that email, message or fill in your booking form?
Apply the same rule, because the channel changes and the question does not. TechCrunch reported that Instinct now gives its assistants their own email addresses for signing up to services and managing accounts. A booking request may therefore arrive from an address that belongs to an agent, not to your customer.
So when an email or web form reads like it was sent on someone's behalf, do what the card says. Take the request. Ask for the customer's own number or email if you do not hold one. Send your confirmation there.
What I Tell Business Owners
What I tell business owners is that this is a people problem that happens to come down a phone line. The line on my business page says it plainly: "The hard part of AI at work is not the technology. It is helping people use it well, safely and every day. That is what an educator does."
In a small business, the person with the least guidance about AI is often the one who answers the phone. Owners spend their attention on which assistant to buy for the office. Meanwhile the receptionist, the service adviser and the Saturday staff are the ones who will speak to an agent first, and they have been told nothing. So they improvise. Some hang up. Some hand over whatever is asked for, because the voice was polite and knew the customer's name.
Neither is a failure of the person. Both are a missing sentence from the owner. If you run your own agent one day, the same thinking applies from the other side, and the Errand Test decides which errands it may run. If AI already answers your customers, the rule for AI in customer service keeps it to what you have written down.
If you run a chamber, a trade association or an industry event, this is a subject members ask about before they know what to call it, and it is ground my business keynotes cover.
What if your business is in the UK?
Then you have a little time, and it is worth using. None of the pages I read lists the UK. Google says its customer calls run only in select regions and names US states. The Gemini help page says it can call only US phone numbers. The reporting on Muse and Instinct describes calls to US businesses.
UK firms are more likely to meet agents first by email and web form. The rules for the companies that build agents are also still being written. The Information Commissioner's Office opened a call for evidence on agentic AI on October 8, 2026, which closes at the end of November 20, 2026, "to inform our thinking on how to apply data protection law to the distinctive capabilities and risks of agentic AI" (ICO, October 2026).
Your own position does not wait for that. You already may not read out one person's details to someone who has not shown they act for them. An agent changes the voice on the line, not that duty. I am not a lawyer, and this is not legal advice. If you hold sensitive records, ask whoever advises you on data protection, and write the answer into your one-page AI policy.
What should you do this week?
A 20-person business can do all of this in an hour.
- Fill in the Front Desk Card. Six lines. Put one name on line five.
- Tell everyone who answers the phone or the inbox. Two minutes at the start of a shift. Give them the opening sentence.
- Check your listing. Hours, services and phone number, identical on your website and your Business Profile.
- Decide about Google's calls. Open the Advanced settings and choose on purpose. Leaving it on by accident and turning it off in a panic are both worse than deciding.
- Write your confirming text. One reusable message with the date, the time, the job and how to change it.
- Log the first five. After five agent contacts, read the log with the team. Change the card where it did not fit.
If your shop runs on walk-ins and bookings, the sector guides for retail and hospitality and trades and construction show where else AI touches the front of the business.
Where to go next
If your front-line staff have never been shown what AI can and cannot be trusted with, that is what my AI training for teams is built for. If you would like to see where the business stands first, the free Business AI Readiness Scorecard takes about five minutes: 12 questions across six areas. Or talk to me about your team.
The agent will ring whether or not you are ready. Decide today what it gets.
Sources and further reading
- About automated calls and texts from Google to your business, Google Business Profile Help, read November 16, 2026 (the page shows no date).
- Ask Gemini to handle your everyday phone calls, Gemini Apps Help, read November 16, 2026 (the page shows no date).
- More advanced AI capabilities are coming to Search, Robby Stein, Google, July 16, 2025.
- Rival AI agents, Instinct and Meta's Muse, both add the ability to make calls, Ivan Mehta, TechCrunch, September 17, 2026.
- Meta testing a 'human concierge' for its new personal AI agent, Muse, Reuters, published by BNN Bloomberg, September 22, 2026.
- New Research: UK small businesses want AI to win back time, not just write copy, Small Business Institute with Alibaba.com, September 17, 2026 (1,000 UK small business owners; fieldwork August 2026).
- Agentic AI Call For Evidence, Information Commissioner's Office, opened October 8, 2026, closes November 20, 2026.
Dan Fitzpatrick is a keynote speaker, Forbes contributor and five-times bestselling author who has trained more than 150,000 people across 30+ countries. About Dan



