Do You Have to Tell Callers They're Talking to an AI in Canada?

If you put an AI receptionist on your business line in Canada, sooner or later someone will ask whether it has to announce that it's an AI. It's a fair question — several US states now require exactly that, and a lot of vendor blog posts treat the answer as a settled yes.

For inbound calls in Canada, the honest answer today is no: no law currently in force requires an AI phone agent to announce itself on a call your customer made to you. But that isn't the same as "say nothing," and the ground is moving. Here's what actually applies, what's coming, and what we'd do in your position.

*This is a plain-language summary, not legal advice. If your business is in a regulated field — health, finance, insurance, legal — check with your own counsel.*

What does not exist (yet): a Canadian AI disclosure law

Canada came close. The Artificial Intelligence and Data Act was part of Bill C-27, and it would have put transparency duties on "high-impact" AI systems. It died on the Order Paper when Parliament was prorogued in January 2025, and the federal government has said it won't bring it back as drafted. There is a voluntary federal code of conduct for generative AI, but voluntary is the operative word.

Provinces have passed narrow AI rules — Ontario, for instance, now requires employers to say when AI is used to screen job applicants — but none of them reach a receptionist answering the phone.

So if a vendor tells you Canadian law requires their agent to announce itself on inbound calls, ask them which statute. There isn't one.

What does apply today

### 1. Recording notice — this is the one people miss

Every AI phone agent captures the call in some form. At minimum it keeps a transcript, because that's how you review what happened afterwards, and many — ours included, if you turn it on — keep the audio as well. Either way you are collecting what the caller said, and federal privacy law expects you to tell them.

The Criminal Code allows a recording when one party to the call consents, and your business is a party — so recording isn't a crime. But PIPEDA, the federal private-sector privacy law, is the stricter layer. The Privacy Commissioner's guidance on recording customer calls expects businesses to tell callers at the start of the call that it's being recorded, to say why, and to offer another way to reach you if they object. A caller who hears the notice and carries on has consented to that purpose.

In practice, this is the familiar "this call may be recorded for quality purposes." Most small businesses that switch from a human receptionist to an AI one never had to think about it before, because nobody was recording. Now something is.

### 2. Quebec: automated decisions

Quebec's Law 25 goes further than the rest of Canada on automation. Since September 2023, a business that makes a decision about someone *based exclusively on automated processing* of their personal information must tell them, and must let them ask for a human to review it.

For most front-desk calls — taking a message, booking an appointment into your calendar, answering "are you open Saturday?" — the agent isn't deciding anything about the caller, and this doesn't bite. It does if you have the agent approve or decline something: financing, eligibility, a claim. If your agent says no to people in Quebec, talk to a lawyer before it does.

### 3. Outbound calls: the CRTC's existing rules

If your agent places calls — follow-ups, reminders, reactivation campaigns — you're in CRTC territory. The Unsolicited Telecommunications Rules have covered "pre-recorded or synthesized" voice messages for years, and an AI voice is synthesized. For sales calls that means express consent before dialing, plus an opening message that says who the call is on behalf of and why, and a way to reach you. None of that is new; it just now applies to software as well as people.

What's coming: the CRTC's 2026 review

In June 2026 the CRTC opened its first full review of the telemarketing rules since 2014. Two of its questions are directly about AI:

• whether the definitions should explicitly cover calls made with synthesized voices and AI, and
• whether callers should have to tell people, *at the start of the call*, that they are not speaking to a live person.

It also asks about AI agents placing calls for purely personal reasons, like booking a restaurant table — a sign the Commission is thinking beyond classic telemarketing. Comments closed in August. No decision has been published yet, and the review is formally about unsolicited calls, not calls your customers make to you. But it's the clearest signal so far of where Canadian rules are heading, and once "tell them it's an AI" becomes a legal requirement for one kind of call, it rarely stays confined to it.

Should you disclose anyway?

We think yes, for reasons that have less to do with law than with your customers.

• You already owe them a recording notice. Adding "you're speaking with our AI assistant" to that line costs nothing.
• People ask. Callers regularly ask "am I talking to a robot?" An agent that dodges the question does more damage to trust than one that said so in the first sentence.
• It sets expectations. A caller who knows it's an AI speaks a little more clearly, is more patient with a pause, and is less likely to feel misled when they ask for something only a person can do.
• It's one sentence. Roughly five seconds at the top of the call. Most callers barely register it.

The one case to think about is if your callers are overwhelmingly repeat customers who already know — a clinic where every patient has heard the AI a dozen times, say. That's a reasonable judgement to make for yourself, which is why we made it a setting rather than a rule.

How we handle it in Voismatic

Voismatic has an AI and recording disclosure setting for your whole business, off by default, under Settings → Calls. Turn it on and every call — inbound and outbound, whichever agent answers — opens with a short line before the greeting. The default is:

> "Just so you know, you're speaking with [your business]'s AI assistant, and this call is recorded."

That last word follows the Record calls setting beside it: with recording off, the default says the call is *transcribed* instead, because telling a caller something is recorded when it isn't is its own kind of misleading.

You can rewrite it in your own words, and it's spoken exactly as written, even when the agent personalises its greeting for a returning caller. That last part matters: a disclosure the AI is allowed to paraphrase is a disclosure you can't be sure was said.

Whatever platform you use, check three things:

1. Can you make it say a disclosure verbatim, on every call, before anything else?
2. Does it say "yes" when a caller asks if it's an AI? Call it and ask.
3. Does the transcript show the disclosure was said? If a complaint ever comes in, "it's turned on in settings" is weaker evidence than the call itself.

The rules in Canada will likely tighten in the next year or two. The businesses that won't need to change anything when they do are the ones already opening every call with a plain, honest sentence.