The CRTC wrote its telemarketing rules for autodialers years before anyone was pitching an AI voice agent. The definitions did not need updating to reach one: a synthesized voice was already built into the rule, which means an AI system placing outbound calls is regulated the same way a recorded robocall always has been.
Key takeaways
An AI voice agent that calls a customer list to offer a renewal, confirm an appointment, or run a survey is not creating a new regulatory category. It is doing, with a generated voice instead of a recording, exactly what the CRTC’s telemarketing rules were written to cover.
The CRTC’s Unsolicited Telecommunications Rules define an Automatic Dialing-Announcing Device as equipment that stores or produces telecommunications numbers in order to convey a pre-recorded or synthesized voice message to a phone number. (CRTC, Unsolicited Telecommunications Rules) A generated voice is a synthesized one under any ordinary reading of that phrase — the rule does not distinguish a recording made once from speech generated fresh for each call.
The ADAD rules apply on top of, not instead of, the National Do Not Call List regime: they bind a telemarketer “whether or not the telemarketing telecommunication is exempt from the National DNCL Rules.” Express consent for an ADAD call has to “clearly evidence the consumer’s authorization” and “include the specific telecommunications number” being called, and it can be withdrawn at any time. (CRTC, Unsolicited Telecommunications Rules) Critically, the rules place the burden of proof on the caller: it is the telemarketer’s job to demonstrate that valid consent existed, not the recipient’s job to prove it did not.
A call that is not trying to sell anything — a service update, an appointment reminder — still has to satisfy a list of conditions under the rules: calling hours restricted to roughly 9:00 a.m. to 9:30 p.m. on weekdays and 10:00 a.m. to 6:00 p.m. on weekends in the recipient’s own local time; an opening message that identifies who the call is on behalf of and how to reach them, repeated if the call runs past sixty seconds; a ban on sequential dialling; and equipment that disconnects within ten seconds of the recipient hanging up. (CRTC, Unsolicited Telecommunications Rules) None of these turn on whether a person or an AI system is running the call.
A phone call is not a “commercial electronic message” under Canada’s anti-spam law, which governs messages sent “to an electronic address.” (CASL, s.6) Voice calling and commercial email or text sit under two separate regimes with two separate consent tests — a business running an AI system across both channels has to clear each one on its own terms, not assume that meeting one satisfies the other. The mechanics of CASL consent for an AI-drafted message are covered in a companion piece on this hub.
The rules also govern what happens after the call. A telemarketer has to keep records in the ordinary course of business, maintained at its regular place of business and readily accessible, and produce them to the CRTC within thirty days of a request; telemarketing registration records specifically have to be kept for three years from creation. (CRTC, Unsolicited Telecommunications Rules) Combined with the rule that the onus of proving valid consent sits on the caller, that means an AI voice programme needs its consent records built into the system from day one — reconstructing them after a complaint is not realistically possible.
The rules distinguish a call that reaches a live person from “voice mail broadcast” — defined as a recorded message delivered directly into a voicemail box without interrupting the recipient’s activities in real time — and state plainly that the Telemarketing Rules do not apply to it. (CRTC, Unsolicited Telecommunications Rules) An AI system built to leave a voicemail rather than to conduct a live or interactive call sits in a genuinely different category under the rules — but the distinction turns on the technical delivery method, not on how “AI” the system marketing describes itself as, so it is worth confirming exactly how a specific vendor’s tool actually delivers a message before relying on the exemption.
A brokerage deploys an AI voice agent to call last year’s clients about a mortgage renewal. Before the campaign runs, the business needs the specific number on file with documented, withdrawable express consent to be called for that purpose; the calling window has to sit inside the permitted local hours for each recipient’s time zone, not the business’s own; the opening line has to identify the brokerage and how to reach it; the system has to disconnect within ten seconds if nobody answers or the line is hung up; and the consent records have to be retrievable inside thirty days if the CRTC ever asks. None of that changes because a generated voice, rather than a person, is speaking the words.
Related: CASL and AI-written email, and disclosing that a customer is talking to an AI system.
How a business designs an AI-driven outbound calling programme around these rules is covered on the AI growth and marketing hub.
No. The rules attach to whoever initiates, causes, or permits the call to be made, and that responsibility is not erased by automating the dialling or the script an AI system reads from.
Non-solicitation ADAD calls are still regulated — they carry their own set of conditions covering calling hours, identification, sequential dialling and disconnection, distinct from the consent rule that applies specifically to telemarketing calls.
It can. The rules treat a message delivered directly into a voicemail box without interrupting the recipient in real time as a separate “voice mail broadcast” category the Telemarketing Rules do not apply to — but that turns on the technical delivery method, and a business should confirm exactly how a specific tool delivers its message before assuming the exemption applies.
Yes, at any time. The rules state plainly that a consumer may withdraw their express consent whenever they choose, and once withdrawn the business needs a way to remove that number from an automated calling programme immediately, not on the next scheduled list refresh.
A short call is enough to walk through consent, timing windows and identification requirements for a specific campaign.
