Can we say our product is "AI-powered"?

“AI-powered” is on almost every startup homepage. Canada has no AI-specific advertising rule, but the Competition Act applies to AI claims like any other claim. In its September 2026 submission on AI transparency, the Competition Bureau said there is no “AI exception” to the law.

What is AI washing?

AI washing means overstating the role or ability of AI in a product: calling a keyword filter “AI”, describing a manual service as automated, or claiming capabilities the model doesn’t have. The legal risk isn’t the word “AI” itself. It’s whether a customer would come away with a false impression about what the product is or does.

Which rules apply?

Rule	What it covers
Competition Act, s. 74.01(1)(a)	Civil reviewable conduct: a representation to the public that is false or misleading in a material respect
Competition Act, s. 74.01(1)(b)	Civil reviewable conduct: a claim about a product’s performance, efficacy or length of life that isn’t based on an adequate and proper test, with the burden of proof on the person making it
Competition Act, s. 52	Criminal offence: knowingly or recklessly making a materially false or misleading representation
Consumer Protection Act, 2002 (Ontario)	False, misleading or deceptive representations to Ontario consumers are unfair practices, and consumers can have remedies
Securities law	Statements to investors, see below

Under the Competition Act, both the literal meaning and the general impression of a claim count, so fine print rarely fixes a misleading headline. On the civil side, penalties for a corporation can reach the greater of $10 million ($15 million for a later order) and three times the benefit from the conduct, or 3% of worldwide gross revenues if the benefit can’t be determined (s. 74.1). Since June 20, 2025, private parties, including competitors, can also apply to the Competition Tribunal with leave.

Ontario has passed a Consumer Protection Act, 2023 to replace the 2002 Act, but as of October 2026 it isn’t in force.

What needs testing before we publish?

Any claim about what the product achieves is a performance claim under s. 74.01(1)(b). For AI products, that usually includes:

• Accuracy rates (“98% accurate”, “fewer errors than manual review”)
• Time or cost savings (“cuts review time by 70%”)
• Outcome claims (“detects fraud”, “catches every risky clause”)
• Capability claims (“understands any language”, “fully autonomous”)

The testing has to exist before the claim is made, has to fit the claim, and should reflect real conditions, not a benchmark that doesn’t match how customers use the product. If the claim is about results for a type of customer, test on that type of data. If the model changes, recheck the claim.

Can we compare ourselves to competitors or to humans?

Yes, if the comparison is accurate and tested. Comparative claims (“more accurate than [competitor]”, “as good as a lawyer”) need testing against the thing you’re comparing with, under fair conditions. Claims that an AI tool replaces a professional carry extra risk. In the US, for example, the Federal Trade Commission’s 2024 action against DoNotPay targeted “robot lawyer” claims that hadn’t been tested, and in 2025 it finalized an order against Workado over a claimed 98% accuracy rate for an AI content detector. US cases don’t decide Canadian law, but they show the kinds of claims regulators look at. If your product touches health, see health claims in advertising.

What about “human-reviewed” claims?

These are representations too, and they cut the other way. If you say every output is reviewed by a person, that has to be true for every output, not a sample. Say what actually happens, for example “a sample of outputs is reviewed weekly”, and keep records showing it. Presenting a chatbot as a person can also mislead. See do I have to tell people my product uses AI and liability for AI chatbots.

What about what we tell investors?

Claims in a pitch deck or offering document are judged by securities law, not just advertising law. Overstating your AI, its revenue impact or the share of work it does can support misrepresentation claims by investors, and if you give investors an offering memorandum, Ontario’s Securities Act gives them statutory rights for misrepresentations in it. The Canadian Securities Administrators flagged AI washing in Staff Notice 11-348. Keep deck claims consistent with your website and your data room. See investor due diligence.

How do we keep a substantiation file?

• List each claim on your site, in ads, sales decks and app store listings.
• Attach the support: test method, dataset, date, model version and results.
• Match wording to evidence. If testing shows 91% on one task, don’t claim “over 90% accurate” generally.
• Qualify clearly and nearby, not in a footnote.
• Review on each model or product change, and retire claims that no longer hold.
• Control AI-generated marketing. If you use AI to write copy, reviews or testimonials, check it before it goes out. Fake or AI-written reviews presented as real are misleading.

What to do next

We draft terms of service and AI product terms that match what your product actually does, at fixed fees on our fees page. For the wider rules, see our AI law tracker and resources for AI startups.

This guide is general information about Ontario and Canadian law, not legal advice for your situation. To check your AI claims before launch, book a free 15-minute consult.