Treadstone Associates
Article · Marketing compliance

Can I say my firm is AI-powered?

Yes — if it is true and the impression it leaves is accurate. The general impression conveyed counts as much as the literal words, and a performance claim you cannot substantiate is the trap.

Treadstone Associates · Updated 2026

Key takeaways

  • • Section 74.01(1)(a) prohibits representations to the public that are false or misleading in a material respect, for the purpose of promoting a business interest.
  • • Paragraph 74.01(1)(b): a performance claim must be based on an adequate and proper test, and “the proof of which lies on the person making the representation”.
  • • Subsection 74.03(5) makes the general impression, not just the literal meaning, the test.
  • • Subsection 74.1(3) provides a due diligence answer. A documented marketing sign-off is what that looks like in practice.

The short answer

Yes, if it is true and the impression it leaves is accurate. The constraint is not a rule about the word “AI” — it is the general prohibition in Section 74.01 of the Competition Act on making a representation to the public that is false or misleading in a material respect, for the purpose of promoting a product or any business interest.

Two features of that provision decide most cases. subsection 74.03(5) requires the court to take into account the general impression conveyed by a representation as well as its literal meaning. And the Competition Bureau explains that material information is information that could influence consumer behaviour — with no need to show that anyone was actually deceived, that they were in Canada, or that the representation was made anywhere the public had access.

Three claims that get professional firms into trouble

1. Capability claims that are literally true and generally misleading. “Our AI reviews every contract” is defensible if a model does in fact pass over every contract. It is a problem if the general impression is that the review is autonomous when a person does the substantive work, or the reverse — that a senior professional reads everything when a model produces the first and only pass. The literal words are not the test on their own.

2. Performance claims with nothing behind them. Paragraph 74.01(1)(b) is the one firms walk into. A representation in the form of a statement, warranty or guarantee of the performance, efficacy or length of life of a product that is not based on an adequate and proper test thereof is reviewable conduct — and the provision says in terms that the proof of which lies on the person making the representation. “Cuts turnaround by half” and “99% accurate” are performance claims. The test has to exist before the claim is published, and it has to be yours.

3. Attribution and association claims. The Bureau’s guidance on the attributes of a supplier is direct: promotional information should not claim the business is associated with, authorized by, or has a relationship with a third party unless it is true. “Certified partner”, “built on” and a vendor logo in your footer are all representations.

Safer phrasing, and why

  • • Risky: “AI-powered document review that catches what humans miss.” Why: a comparative performance claim needing an adequate and proper test you can produce.
  • • Defensible: “We run an automated consistency pass over every draft before a partner reviews it.” Why: it describes a process you either do or do not do.
  • • Risky: “Save 40% on your compliance costs.” Why: a quantified outcome claim about the customer’s result.
  • • Defensible: “Fixed monthly fee, quoted before we start.” Why: a term of business, verifiable from your own engagement letters.
  • • Risky: “Powered by [vendor], an official partner.” Why: an attribution claim; if the agreement does not say partner, do not say partner.
  • • Defensible: “We use [vendor] under a commercial licence.” Why: accurate and unglamorous.

What the exposure looks like

Section 74.1 sets out what a court may order on a finding of reviewable conduct: an order not to engage in the conduct, publication of a corrective notice, and an administrative monetary penalty. For an individual the ceiling is the greater of $750,000 (rising to $1,000,000 for each subsequent order) and three times the value of the benefit derived. For a corporation it is the greater of $10,000,000 (rising to $15,000,000 for each subsequent order) and three times the benefit, or where that cannot reasonably be determined, 3% of annual worldwide gross revenues. An order not to engage in the conduct applies for ten years unless the court specifies less.

There is a defence worth knowing about. Subsection 74.1(3) provides that no order may be made under paragraphs (1)(b), (c) or (d) if the person establishes that they exercised due diligence to prevent the reviewable conduct from occurring. A documented sign-off step on marketing copy is exactly the kind of thing that supports it. A model generating claims straight onto a live page is exactly the kind of thing that does not.

The AI-specific failure: the model writes the claim

The realistic risk for a professional firm is not a deliberate exaggeration. It is a generative tool asked to “write a homepage for an AI-enabled accounting practice” producing confident, conventional marketing language — faster, more accurate, fewer errors — which a busy partner approves because it sounds like everyone else’s site. Nobody decided to make a performance claim. One got made anyway, and section 74.01(1)(b) does not care how it arrived.

The fix is procedural. Keep a list of claim types that require evidence before publication, and make the marketing sign-off explicit about which category each sentence falls into. Beyond the statute, the Canadian Code of Advertising Standards is the self-regulatory baseline, and Canada’s Voluntary Code of Conduct on advanced generative AI systems is the closest thing to a stated federal expectation about transparency in AI deployment.

Worked example (illustrative)

A Toronto bookkeeping firm drafts a new homepage with a generative tool. Three lines get flagged in review. “AI-powered reconciliation that never misses a transaction” becomes “every transaction is matched automatically and exceptions are reviewed by a bookkeeper”. “Clients save hours every week” is deleted, because nobody had measured it. “Official Xero partner” is checked against the actual agreement and kept, because it was accurate.

The whole review took twenty minutes and produced a file note recording who approved what and on what basis. That file note is the due diligence evidence, and it costs nothing to create at the time and a great deal to reconstruct later.

Where this sits in the firm

This is your firm’s own marketing. The rules for sending it are in the CASL article. For the underlying law see Treadstone’s sister firm on misleading advertising, comparative advertising and testimonials and endorsements.

Questions we get asked

Can I repeat a vendor’s statistic about their own product?
If you publish it to promote your business, you have made the representation, and paragraph 74.01(1)(b) puts the proof on the person making it. Attribute it, or better, describe what the tool does for your clients rather than what a vendor says it does in general.

Is “AI-powered” misleading if it is really a rules engine?
It might be, on the general impression test. If a plain description of the mechanism would disappoint the reader relative to the label, that gap is the risk.

Do I have to disclose AI use to clients?
The Competition Act does not require it; your professional obligations and your engagement terms are a separate question, and some courts have published expectations about AI use in materials filed with them — the Federal Court’s AI page is the clearest Canadian example.

Does a disclaimer fix an overstated headline?
Not reliably. Subsection 74.03(5) requires the general impression to be taken into account, and a headline that carries the impression is not cured by small print that contradicts it.

Have your claims reviewed before they go live.

A 30-minute call is enough to tell you whether AI pays for itself here.