Yes, for most requests, and on your terms. AI assistants already place phone calls to businesses for the people they work for, and a blanket "we don't talk to bots" policy mostly pushes the same customer back to you through a slower path. The useful rule is simpler: answer the AI caller, give it exactly what you would give an unverified human caller, and make anything account-specific depend on the customer proving it is them and that they want it done. The assistant can carry the request. It cannot carry the customer's identity.

That is also how a voice AI concierge like Lorikeet's handles the other end of the line: it answers on local numbers, verifies identity before account actions, takes those actions through your systems, and hands over to your team with the full conversation when it should. Lorikeet publishes this guide and sells AI customer support across voice, chat and email, so weigh our product sections accordingly. The legal points below are general information, not legal advice; confirm the rules for your business with counsel.

Key takeaways

• AI assistants calling businesses is not hypothetical. Google's Gemini app can call businesses for eligible US users, and it opens every call by saying it is an AI assistant from Google.

• Separate two questions: who the customer is, and whether they authorised this caller to do this. A caller that knows the account number has answered neither.

• Serve information and booking requests freely. Require the customer, not the assistant, to complete verification for account access, and keep high-risk actions for the customer or a human agent.

• The FCC's February 2024 ruling on AI voices is about calls placed with artificial voices under the TCPA. It governs the caller, which matters when your own AI calls customers, not when you answer.

• Recording rules still apply to every call. Keep your disclosure, and check state and national rules with counsel.

AI assistants are already calling businesses

Google showed the idea in 2018. Its research post on Google Duplex described a system for completing specific tasks over the phone, such as restaurant bookings and hair salon appointments, with the user not taking part in the call, and said Google wanted to be clear about the intent of the call so businesses understood the context.

It has since become a mainstream feature. In July 2025, Google said Search could call local businesses to get pricing and availability on a user's behalf, rolling out to all Search users in the US. Google's help page for the Gemini app's phone calls now says Gemini can get business hours and service quotes, book, confirm and manage appointments and reservations, and navigate phone menus and wait on hold. It also says Gemini starts every call by disclosing that it is an AI assistant from Google, and that it cannot make payments or relay sensitive personal information such as social security numbers, financial information, passwords or health information.

Google's Business Profile help describes the business side: automated calls to check details like hours or handle customer requests such as booking appointments, calls that are monitored and recorded for quality assurance, and an opt-out by telling the caller to stop or changing Business Profile settings. Other assistants are following. For the patterns to watch for in your own queues, see signs AI agents are already contacting your support team.

When to serve an AI caller and when to decline

The conventional instinct is to treat AI callers as a fraud problem. Some are. Most are a customer who did not want to wait on hold, and the request tells you more about the risk than the voice does. Sort by what is being asked.

Declining is a legitimate answer for the bottom rows. Do it plainly, say what the customer can do instead, and log it. A vague refusal just produces a second call. Our guide to blocking versus serving AI agents covers the trade-offs across channels.

Identity and authority are different questions

Traditional phone verification assumes the person on the line is the customer. With an AI caller, that assumption breaks in two places.

First, identity. Whatever the assistant knows, it can repeat: date of birth, postcode, the last four digits of a card. Knowledge questions were already weak; NIST's current identity proofing guidance says knowledge-based verification shall not be used for identity verification. A voice is not proof either. FinCEN's alert on deepfake media reported criminals using generative AI to create falsified documents, photographs and videos to get past customer identification and verification.

Second, authority. Even a genuine assistant acting for a genuine customer may have been asked to do one thing and attempt another. An account number finds an account. It does not prove the customer asked for this action.

The practical fix is to send verification to the customer, not through the caller: an approval prompt in your app, a passkey, or a call back to the number on file. If the assistant cannot complete the step alone, the control is working. Be careful with texted one-time codes, because many assistants can read their user's messages and relay the code. Our deeper guide covers how to verify identity when an AI agent contacts support.

Consent, disclosure and recording: what actually applies

This is where most commentary gets sloppy, so precisely:

The FCC's AI voice ruling covers the caller. In February 2024 the FCC issued a declaratory ruling confirming that the TCPA applies to AI technologies that generate human voices: calls using AI-generated voices count as "artificial or prerecorded voice" calls, which need the called party's prior express consent unless an emergency purpose or exemption applies. It is aimed at robocalls placed to consumers. When an assistant calls your contact center, the TCPA obligations sit with whoever places the call, not with you for answering. The ruling matters to you when your own AI calls customers, for reminders, follow-ups or callbacks.

Recording consent still applies to you. Under US federal law, 18 U.S.C. 2511(2)(d) allows a party to a call, or someone with one party's prior consent, to record it. State and national laws vary and some are stricter, so keep your recording disclosure at the start of every call, AI caller or not. Note that assistants record too: Gemini, for example, tells the business the call is on a recorded line.

Disclosure is a norm worth requiring. Reputable assistants say what they are and who they act for at the start of the call. Make it part of your policy: if a caller says it is an assistant, ask whose behalf it is calling on and what it has been asked to do, and record both in the ticket.

Healthcare teams should keep AI callers to operational requests such as scheduling and directions, and confirm with counsel how their privacy obligations apply before discussing anything about a patient with a third party.

Handling AI callers in the IVR and with live agents

Assistants can navigate phone menus and wait on hold, so your IVR is already serving them. A few changes make it serve them well:

• Accept both speech and keypad input, and avoid menus that only work by pressing a key after a long prompt.

• Route requests by intent, not by caller type. An AI caller asking for opening hours belongs in the same self-service path as a human asking the same thing.

• Tag the ticket when a caller identifies itself as an assistant, so you can measure volume and spot abuse.

• Give live agents a short script: confirm who the assistant represents, what it has been asked to do, and what the customer must do to verify.

• Do not penalise agents' handle time for AI calls while you learn what they look like.

• If you decline, say so clearly. Google's automated callers, for one, stop when a business asks them to.

The background on running voice alongside other channels is in our AI voice and SMS support guide.

Platforms and approaches worth considering

1. A voice AI concierge on your lines: Lorikeet

Lorikeet Voice answers and places calls on local numbers in the US, UK and Australia, verifies identity, takes actions through your APIs during the call, and escalates to your team with the full conversation. Identity checks and account actions are deterministic steps the AI can call but not alter, and it works alongside telephony such as Amazon Connect, Twilio, Five9 and Dialpad. AI callers reach the same concierge as customers, with the same guardrails, identity verification and action policies, and no more authority than the customer has. One limit to know: Lorikeet does not hold PCI DSS certification, so card details must go through your existing PCI-compliant system.

2. Google Business Profile controls

If Google's automated calls are a large share of your AI calls, the Business Profile settings let a verified business turn them off, and Google says you may still get manual calls to confirm your information. Turning them off also removes a route customers use to book with you, so decide deliberately.

3. A written AI-caller policy for your existing team

No software at all: a one-page policy using the serve or decline table above, a verification method that reaches the customer directly, and a ticket tag. This is the minimum every contact center should have, whatever else it runs.

4. A non-phone path for agents

Phone is the most expensive way for an assistant to ask you a question. Publishing a documented endpoint or an agent-readable page lets assistants get answers and complete simple tasks without a call. Our explainer on B2A (business to agent) covers how that channel works.

Checklist for your contact center

• Decide which request types AI callers can complete, which need customer verification, and which need a person.

• Replace knowledge questions with verification the customer completes on their own device or a call back to the number on file.

• Ask every AI caller whom it represents and what it has been asked to do, and record both.

• Keep your recording disclosure on every call, and check state and national rules with counsel.

• Check consent before any outbound call your own AI places, under the TCPA and your local rules.

• Tag AI-caller tickets and review a sample weekly for abuse, failed verification and dead ends.

• Make sure the IVR accepts speech and keypad input and routes by intent.

• Write the decline wording and the alternative you offer.

Next steps

If AI callers are already showing up in your queues and you want a voice concierge that applies your verification rules on every call, get a demo or start a 30-day free trial and build your first concierge with Coach, the setup agent.