This question almost never arrives as a legal question. It arrives as a disagreement. One person on the marketing team drafts with AI and says nothing. Another thinks every post that AI touched should carry a label. Both of them are trying to do the right thing, and nobody has written down which one is right.
The good news is that the rules are narrower and clearer than the argument suggests, and the part they leave open is yours to decide.
What follows is the lay of the land as of November 2026, with the sources linked. It is not legal advice. If you work in a regulated field or run large paid campaigns, have a lawyer read your final policy.
Is there a law that says I have to label AI content?
In Canada, not for ordinary marketing. There is no federal rule that says a newsletter, a caption or a product description has to announce that AI helped write it.
What Canada does have is a rule about the result. The Competition Act prohibits marketing claims that are false or misleading in a material respect, and the Competition Bureau is plain about how that gets judged: by the literal meaning and by the general impression the ad makes. Material means it could influence what someone buys or does.
Notice what that test does not ask. It does not ask what tool you used. A misleading photo is misleading whether it came from a camera, an editing suite or an image generator. An accurate one is fine from any of them.
So the working question is not “did AI touch this?” It is “does this leave someone believing something that isn’t true, and would they care?”
When does AI use cross into misleading?
When the customer would reasonably take the content as real, and the realness is part of why they’d act.
- A picture of a result your product did not produce.
- A testimonial, review or endorsement from a person who does not exist.
- A photo of your place, your team or your event that shows something a visitor will not find when they arrive.
- A real person’s face or voice saying something they never said.
Canada’s advertising self-regulator has put this in writing for influencer content. Ad Standards’ disclosure guidelines, updated in October 2025, say that when an image showing a product’s benefits or attributes is generated by or significantly altered using AI, that should be disclosed, and they suggest tags like #MadeWithAI. They add that content made with filters or AI still has to be truthful about benefits, results and attributes. These are industry guidelines rather than legislation, but they are a fair picture of where the expectation sits.
What doesn’t need a label?
Most of what a small team actually uses AI for.
Drafting a caption you then edit and approve. Tidying a paragraph. Brainstorming subject lines. Summarizing your own notes into a newsletter intro. Resizing an image or cleaning up its background. In all of these, a person on your team read the thing, changed it, and put the business’s name on it. The claims in it are yours, and you stand behind them. That is the same accountability you had when a freelancer or an agency wrote the first draft, and nobody labelled those.
The platforms draw the line in about the same place. YouTube requires disclosure when realistic content makes a real person appear to say or do something they didn’t, alters footage of a real event or place, or shows a realistic scene that didn’t happen. It does not require it for production help like scripts, outlines, titles and thumbnails, or for minor aesthetic edits.
Do the social platforms have their own rules?
Yes, and they are the rules most likely to actually reach you, because they apply the moment you upload.
YouTube’s are described above. Meta and TikTok have their own versions for realistic AI-generated video, audio and images, and Meta publishes its guidance for businesses. The details shift, so check the current wording for each platform you post on, especially before running ads. The pattern across all of them is consistent: realistic and could be mistaken for real means label it; assisted, edited or obviously not real means you generally don’t have to.
What if some of my audience is in Europe?
Then one more rule is worth knowing. The transparency obligations in Article 50 of the EU AI Act began applying on 2 August 2026. For a business using AI tools, the relevant part is that deepfakes (realistic AI-generated or manipulated images, audio or video) have to be disclosed as such. AI-generated text published to inform the public on matters of public interest also has to be disclosed, unless it has been through human review and someone holds editorial responsibility for it.
For most small businesses this lands in the same spot as everything above. Reviewed, edited marketing copy with your name on it is not the target. Realistic synthetic media is.
So what should our own rule be?
This is the part the law leaves to you, and it is where the disagreement on your team gets settled. Legal minimum and your own standard are two different things, and a business built on trust will often choose to go further than it has to. That is a values decision, not a compliance one, and it is a good decision to make on purpose.
A rule that fits on one page:
- Always disclose realistic AI images, video or audio of people, places, products or results. If someone could mistake it for a photograph of the real thing, say so.
- Never use AI for testimonials, reviews, or the words or likeness of a real person without their say-so.
- No label needed for AI-assisted drafting and editing, as long as a named person reviews it and the business stands behind every claim.
- A person approves everything before it goes out. This is the rule that makes rule 3 honest.
- If someone asks, tell them. No customer should ever get a vague answer to “did you use AI for this?”
Then add whatever your values call for on top. Some organizations decide that anything showing the people they serve will always be real photography, with consent. Others go the opposite way and choose clearly labelled illustration specifically because it protects people’s privacy. Both are sound. What matters is that the choice is written down, so two people in the same role aren’t making it differently.
If you don’t have an AI policy yet, our template has a place for exactly this. And if the harder question on your team is about images in particular, we’ve written about where AI images help and where they cost you.
Won’t labelling make people trust us less?
It depends what you’re labelling. A line saying “AI helped draft this email” tells the reader nothing useful and invites them to wonder what else is going on. A clear note that an image is an illustration, on a picture that might otherwise be taken for a photo, does the opposite. It shows you thought about what they’d assume.
The thing that actually damages trust is the surprise: finding out later that something you took as real was not. Disclose where there would be a surprise. Skip the label where there wouldn’t.
Where to start this week
Get the two people who disagree into the same room for half an hour. Pull up the last ten things your business published and sort them into three piles: real, AI-assisted and reviewed, and AI-generated and realistic. The third pile is usually tiny or empty, and it is the only one that needs a label.
Then write the five lines above in your own words and put them where the team works.
If you’d like help turning that into a policy your whole team can stand behind, book a free readiness call. Twenty minutes, no pitch. The disclosure question is usually the first sign a team is ready for a proper conversation about how it wants to use AI.