What to Do When AI Agents Call Your Business: A Small Business Playbook

Treat them like any caller with a job to do. Let your team answer simple questions about hours, wait times, prices and bookings from a short approved script, never give out payment details or customer information, and publish those facts online so agents call less. If the calls cost real staff time, turn off Google's automated calls.

Most of the AI-and-phones conversation is about putting an AI on your own line. A newer question is the reverse: what happens when your customer's AI calls you. A voice says it's "an assistant, calling on behalf of my client," asks what your cheapest haircut costs or whether you have a table for four, and your employee has to decide on the spot whether to help it, hang up, or call a manager.

Why are AI agents calling my business?

Because a customer, or a company collecting data, asked an AI to make the call for them. On October 29, 2026, the San Francisco Standard reported that personal AI agents, naming Meta's Muse, Instinct and OpenAI's dots, can be set up with the right permissions to make calls and send messages to book dinner reservations, dental appointments and car maintenance. An employee at one e-bike shop told the paper he was answering one or two of these calls a week.

The calls come from a few kinds of sources:

• Search and review platforms. Google's Business Profile help center says that when customers use Google Search or Maps, Google may call your business on their behalf to book an appointment, check a restaurant's wait time, or confirm the price and availability of a product or service. On October 20, 2026, Yelp announced Check for Me, an AI phone agent that calls a restaurant, asks the current wait time and texts the diner the answer. Yelp says it's rolling out over the coming months on eligible U.S. restaurant pages on iOS.
• Personal assistants. Agents that work for one person, such as the ones the SF Standard named, calling to book, compare or ask questions.
• Data collectors. Companies building maps or directories that call businesses to gather details. The SF Standard story centers on one: a startup called Voygr, which placed calls to local businesses during the city's Tech Week to test its software.

How can you tell you're talking to an AI agent?

Usually it tells you, or it sounds like a very polite form being read aloud. Google says its automated calls are monitored and recorded, and a Google spokesperson told The Verge that each call starts by announcing that it's an automated system calling from Google on behalf of a user. Other agents are less clear. The e-bike shop employee in the SF Standard story said the agents calling him don't say which company built them.

Common signs: a scripted opening ("I'm an assistant calling on behalf of a customer"), one narrow question, quick replies with an odd rhythm, and trouble with anything off-script, like "we only take walk-ins after 8."

A rule worth teaching everyone who answers your phone: a legitimate agent never needs your passwords, card details or another customer's information (our guide on customer data in AI tools covers why). Google states plainly that it will never ask for payment, credit card information or account passwords, and that a caller asking for money to keep your business on Google is not Google.

Should you answer AI agent calls, or hang up?

Answer the ones that bring you a customer, and keep them short. Voygr's own test, as reported by the SF Standard, is the best real-world look so far. Across 848 calls asking simple questions like the price of a haircut or whether a dentist was taking new patients, about 1 in 3 businesses stayed on the line after hearing it was an AI agent. Veterinary and dental offices were the most likely to keep talking; dry cleaners and bars were the most likely to hang up. These are a startup's results from one city over a few hours, so read them as a signal, not a national benchmark.

Owners in the same story split. One bike shop owner said agents try "to get between business and the customer." The COO of a San Francisco restaurant group said that, as far as she knew, no AI agent had successfully made a reservation at her restaurants. The agents got stuck at payment, tried to book at the last minute, repeated themselves and misunderstood staff. Her bigger worry was agent calls tying staff to the phone and cutting into time with real guests.

That's the tradeoff: a quick wait-time answer may send a party your way, while a long back-and-forth with a confused bot during the rush costs more than it earns.

What should your team say to an AI agent?

Give your staff three simple choices: serve, redirect or refuse. Write them on one page next to the phone.

A one-line script helps: "Happy to answer one quick question. For anything else, please book at our website or leave a number and we'll call back."

What happens if your own AI receptionist answers?

Then it's AI talking to AI, and you need to check that it works. In Voygr's test, about 1 in 25 of the businesses reached already had their own AI answering calls. TechCrunch's coverage of Yelp's launch flagged that a user's agent may end up talking to a restaurant's agent, and that the conversation might break down.

If you use an AI receptionist, test it yourself: call in with an assistant app and ask for a wait time, a price and a booking. Make sure it answers simple facts from your approved list, refuses payment and private data, and sends anything unusual to a person. Our guide on whether a small business should let AI answer the phone covers disclosure, handoffs and what to measure.

How do you get fewer AI agent calls without losing customers?

Put the answers agents are calling for where they can find them without calling. Most agent calls ask the same handful of questions: hours, wait times, prices, availability and booking. If those facts are current on your website and listings, many agents can read them instead.

• Keep your Google Business Profile current. Hours, services, price ranges and booking links. Google says it calls when it needs to check business hours or confirm details.
• Share product data with Google if you sell in-store items. Google says that if you've already shared in-store product details through Google Merchant Center, it will use that information and won't call your business to ask about inventory.
• Publish prices or ranges on your website in plain text, along with your service area, booking rules and walk-in policy.
• Offer online booking. Google says that if you work with an online booking provider, it may book through that partner instead of calling.
• Know your opt-out, and its cost. Google says you can opt out of its automated calls by telling the caller or by changing your Business Profile settings, and that you may still get occasional manual calls to confirm details. Opting out also means Google can't make those booking and price calls for customers who asked it to. If your phone can't handle them, opt out; if you can, the calls are customers raising their hand.

What should you track?

For two weeks, have whoever answers the phone tally agent calls: what they asked, how long they took, and whether any led to a booking or visit. Then decide:

• If agent calls are rare and quick, keep answering and move on.
• If they're frequent and simple, publish the facts they keep asking for and check the count again in a month.
• If they're frequent and wasteful, tighten the script, opt out where you can, and consider routing them to voicemail or a recorded message with your website.

Count first, decide second. The same approach shows up in our guide to answering web inquiries quickly: you can't fix the leak you haven't measured.

How does Caprock Agility help?

We don't sell phone software. We help you decide how calls, inquiries and bookings should move through your business, people and AI included.

Our AI Readiness Audit ($2,500, 3–5 business days) maps how customer contact actually reaches you, including calls, forms and agents, and shows which fix comes first. If you want a working answer-and-book process, EDGE Starter (starting at $8,500, 4–6 weeks) builds one end to end, with scripts, data rules, staff training and a Day 30 outcomes review, on accounts your business controls.

You can compare every option on our services and pricing page.