AI outbound calling is when an AI voice agent places phone calls for a business, rather than answering them. Used legitimately, it rings people who asked to hear from you: callbacks a customer requested, appointment reminders and confirmations, and past customers who have agreed to be contacted. In the United States the Federal Communications Commission treats AI-generated voices as “artificial” voices under the Telephone Consumer Protection Act, so consent rules apply, and Canada has its own Do Not Call and telemarketing rules.
This is general information for business owners, drawn from the regulators’ own pages as at November 2026. It is not legal advice. Rules change and depend on who you call and why, so have a lawyer in the country you are calling check your plan before an AI makes calls for you. CentriWeb builds and looks after voice agents for clients, on platforms of this kind, so weigh our view with that in mind.
What AI outbound calling is used for, legitimately
The calls that work, and that stay on the right side of the rules, share one feature: the person being called expects to hear from you.
- Requested callbacks. A customer fills in a form or leaves a message after hours and ticks a box asking to be called. The AI rings within minutes, answers questions from your own information and books a time.
- Appointment reminders and confirmations. A call the day before asking the customer to confirm, move or cancel, with the change written straight into your calendar.
- Follow-up on an open job. Checking whether a customer received a quote and has questions, where they asked you to follow up.
- Reactivating past customers who have consented. A service reminder or a seasonal check-in to people who agreed to be contacted that way.
- Overflow on a busy day. Returning missed calls when staff cannot get to them.
What we would not use it for: cold calling lists of people who have never dealt with you. That is where the rules are strictest in every country covered here, and where an AI voice is least welcome. Our guide to the AI appointment setter covers the booking side in more detail.
How AI outbound calling works
An outbound voice agent is the same technology as an inbound AI receptionist, pointed the other way. Something triggers the call (a form, a calendar event, a CRM stage), the platform dials, the agent speaks and listens, and the result is written back to your systems. Platforms charge by the minute and some add a small fee per dial: Retell AI’s pricing page, for example, lists batch calling at an extra $0.005 a dial and branded caller ID at $0.10 per outbound call, as at November 2026. Check their page for current prices.
The parts that decide whether outbound calls are lawful are not in the voice. They are the consent record, the do-not-call checks, the calling hours, what the agent says first and how it handles “stop calling me”. Those have to be built into the system before it dials anyone.
AI outbound calling rules in the United States
The FCC’s position on AI voices
The FCC’s declaratory ruling FCC 24-17, adopted on 2 February 2024 and released on 8 February 2024, says the Telephone Consumer Protection Act’s limits on calls using an “artificial or prerecorded voice” cover current AI technologies that generate human voices, including voice cloning. Callers using these technologies must have the prior express consent of the person called, unless an emergency purpose or an exemption applies, and must give the identification and opt-out information the rules require. The ruling also rejects a carve-out for technology that claims to be the equivalent of a live agent.
Consent and the main call rules
The FCC’s regulation, 47 CFR 64.1200, sets out the detail. In summary:
| Rule | What it requires |
|---|---|
| Consent | Prior express consent for artificial or prerecorded voice calls to mobiles; prior express written consent where the call advertises or telemarkets |
| Written consent | A signed written agreement naming the number, stating that signing is not a condition of purchase |
| Identification | Say at the start of the message who is responsible for the call, and give a callback number |
| Opt-out | For telemarketing, an automated opt-out offered within two seconds of the identification |
| Calling hours | No telephone solicitations to residential subscribers before 8 a.m. or after 9 p.m., the called person’s local time |
| Do not call | No solicitations to numbers on the National Do Not Call Registry, using a version of the registry no more than 31 days old |
| Internal list | A written policy, trained staff, and do-not-call requests honoured within a reasonable time of no more than 30 days, kept for five years |
The regulation also allows some non-marketing prerecorded calls to residential lines without consent, within limits of generally three calls in any 30 days. The details matter, so read the section itself rather than relying on a summary.
The Federal Trade Commission’s rule
The FTC’s guide to the Telemarketing Sales Rule adds that outbound telemarketing calls delivering a prerecorded message need the recipient’s prior signed written agreement, whether or not the number is on the registry. An established business relationship (a purchase in the last 18 months or an enquiry in the last three) lets a seller make live calls to a registered number, but the FTC says it does not permit prerecorded message calls. It lists civil penalties of up to $53,088 per violation for calling numbers on the registry or a company’s own do-not-call list.
Our guide to TCPA rules for AI calls in the United States goes through each rule in more depth.
AI outbound calling rules in Canada
In Canada, the Canadian Radio-television and Telecommunications Commission (CRTC) enforces the National Do Not Call List (DNCL) and the wider Unsolicited Telecommunications Rules. The National DNCL telemarketer questions say:
- Any organisation on whose behalf telemarketing is done must register with the National DNCL Operator, even if its calls are exempt.
- You may not make telemarketing calls to numbers that have been on the list for more than 31 days, unless an exemption applies.
- The rules apply to calls made from outside Canada.
- Appointment reminders, payment collection, product recalls and pure market research are generally not telemarketing, but the rules for automatic dialing-announcing devices may still apply to them.
The DNCL exemptions page lists calls to people who have bought, leased or rented from you in the last 18 months, who asked about a product or service in the last six months, or who gave express consent in writing, online or in a recorded call. Calls to businesses are exempt from the DNCL rules. Even for exempt calls, you must identify on whose behalf you are calling at the start, keep an internal do-not-call list, add people within 14 days of their request and keep them on it for three years.
The CRTC’s separate rules for automatic dialing-announcing devices, which cover automated and prerecorded calls, set further conditions. We could not open the CRTC’s own rules page to confirm their current wording, so read them on the CRTC website or with a Canadian lawyer before an AI calls anyone in Canada.
Australia, briefly
Australia has its own rules for telemarketing and outbound calls. Assume they apply to an AI caller as they would to a person on the phone for your business, and check them before you dial. Our guide to AI cold calling rules in Australia covers the detail from the regulator’s own pages.
What a well-run outbound call looks like
As an example: a homeowner fills in your website form at 9 p.m. and ticks “please call me back”, with wording that says the call may come from an automated assistant. The consent, the wording and the time are saved in the CRM. At 9 a.m. their time the next morning, the agent rings, says the name of your business and that it is an AI assistant returning their enquiry, answers two questions from your price guide and books a site visit in the real calendar. When the homeowner says they would rather speak to the owner, it passes the call through. If they had said “don’t call again”, the number would have gone straight onto your internal list.
The same pattern works for reminders and for past customers. For reminders by text as well as voice, see our guide to the appointment reminder text. For bringing back past customers, our guide to a database reactivation campaign covers consent and channel choice.
How to set up AI outbound calling so it stays within the rules
- Record consent where it is given. The form, the wording the customer agreed to, the date and the number, stored against the contact in your CRM.
- Check do-not-call lists before every campaign and keep your own internal list in the same system that dials.
- Set calling hours by the customer’s time zone, not yours, and keep the AI inside them.
- Open every call with who you are and why you are calling. Say early that the caller is an AI where the law or your own standard requires it.
- Make “stop” work. Any request not to be called should end the call politely and update the list straight away.
- Hand over to a person when the customer asks or the call goes beyond what the agent should handle.
- Keep the call records for as long as the rules in each country require, and check the recording laws there too.
When it is worth doing
AI outbound calling pays off when there is a steady flow of calls people expect and staff cannot make in time: callbacks after hours, reminders that cut no-shows, service follow-ups that bring past customers back. Where the alternative is cold calling, the legal risk and the reaction from the people called usually outweigh the saving. If you have the first kind of work, our AI phone and chat agents service covers building outbound calls with the consent and opt-out records built in.