Fabricated reviews and testimonials were a problem long before a language model could write a hundred of them in a minute. The Competition Bureau already has a name for the practice — astroturfing — and a generative tool does not create a new legal question so much as make an old one faster to get wrong.
Key takeaways
A generative AI tool can produce a convincing, specific-sounding customer review in seconds, complete with a plausible name and a believable complaint or compliment. That capability does not change what Canadian law asks about a published review — it just lowers the cost of doing the thing the law already prohibits.
Section 74.02 sets out a specific test for publishing a testimonial or a claim that someone tested a product: the third party must have “previously published the testimonial or represented that he or she has made the test,” and the person publishing it must have “secured in writing the third party’s approval of the testimonial… as well as permission to publish or make it.” The published version must also “accord with” what the third party actually said. (Competition Bureau, use of tests or testimonials)
The penalties for getting this wrong are not nominal. For a corporation, a first violation carries the greater of $10 million (“$15 million for each subsequent violation”) or three times the benefit derived from the conduct, or 3% of the corporation’s annual worldwide gross revenue if the benefit cannot reasonably be determined. (Competition Bureau, use of tests or testimonials)
The Bureau’s own Deceptive Marketing Practices Digest defines astroturfing as “the practice of creating commercial representations that masquerade as the authentic experiences and opinions of impartial consumers, such as fake consumer reviews and testimonials,” and warns that unchecked, it “will seriously erode consumer confidence in the authenticity of online reviews, at a cost to both customers and business.” (Competition Bureau, Deceptive Marketing Practices Digest, Volume 1)
That guidance predates generative AI and was written about employees posting reviews or reputation-management firms paying third parties to post them. The mechanism a business used to fabricate the review has never been the question; the question has always been whether the general impression the review creates — an authentic, impartial customer’s experience — is true.
The general impression test applies specifically to testimonials, alongside sections 52, 52.01 and 74.01. (Competition Bureau, the general impression test) A review generated by an AI tool and published as though a real customer wrote it creates exactly the same false general impression as a paid human-written fake — the legal analysis does not change because a model produced the sentences instead of a person. What changes is how much of it a business can produce, and how quickly a directory or review platform can fill up with it.
The Bureau’s guidance also flags the importance of disclosing a “material connection” between a reviewer and the advertiser, even where the review itself is honest — because “consumers highly value the independence of third-party opinions.” (Competition Bureau, Deceptive Marketing Practices Digest, Volume 1) An entirely fabricated review has no material connection to disclose — it has no underlying customer at all, which is a more serious version of the same problem.
The general impression test is not limited to whether a single quoted sentence is real — it also reaches the impression an aggregate rating creates. The Bureau’s own list of failure patterns includes exactly this shape: “The marketing message is literally or technically true but creates a false or misleading impression. For example, the results of a product test may not be significant, but the marketing message makes it seem like they are.” (Competition Bureau, the general impression test) A handful of AI-generated five-star reviews mixed into a small number of genuine ones can push a displayed average in a way that is “literally” an average of the reviews shown, while creating a false impression of how a typical real customer actually experiences the business.
None of this makes AI itself the problem. A tool used to detect unusual review patterns — a cluster of accounts posting in a short window, near-identical phrasing across supposedly unrelated reviewers — is doing the opposite of astroturfing: helping a business or a platform find and remove exactly the kind of fabricated content section 74.02 and the general impression test are aimed at. The legal question in every case is the same one: does the published material create the general impression of an authentic, impartial customer experience, and is that impression true.
A business asks a generative AI tool to draft a batch of “customer” reviews to fill out a new directory listing, then publishes them attributed to invented names. No section 74.02 defence is available, because there is no third party who made a testimonial to secure approval from in the first place — the entire premise the section is built around does not exist here. The conduct instead falls under the plainer prohibition in section 74.01(1)(a): a representation, by any means whatever, that is false or misleading because it creates the general impression of authentic, impartial customer experiences that never happened. If the fabricated reviews also lift the business’s displayed star rating, that rating is now its own separate, compounding version of the same problem.
Related: AI claims and the Competition Act, and what “AI washing” means.
How a business builds a review and reputation programme that stays inside these rules is covered on the AI growth and marketing hub.
Editing a genuine review for clarity while preserving what the customer actually meant is a different act from inventing the substance of the review or attributing invented opinions to a real or fictional customer. The touchstone under the general impression test is whether the published version still reflects an authentic, impartial experience.
A label describing how a genuine review was processed is a different thing from a label curing a fabricated one — disclosure does not turn an invented opinion into a real customer’s experience, and no Canadian statute currently requires an AI-content label in the first place.
The Act does not set a minimum threshold before a fabricated review counts as a misrepresentation, and even a small number can distort a displayed aggregate rating — the general impression test looks at the impression actually created, not at what share of the underlying content is genuine.
A short call is enough to walk through what a specific review workflow would need to change.
