When AI Invents Facts, Whose Fault Is It?

In February, 2024, Air Canada made an unusual argument before a Canadian tribunal: Its website chatbot, which gave a grieving customer incorrect information about bereavement fares, was responsible for its own mistakes.

The tribunal wasn’t convinced. Tribunal member Christopher Rivers called it “a remarkable submission” and wrote that Air Canada was responsible for all the information on its website, whether it came from a static page or a chatbot. The airline was ordered to pay damages for failing to take reasonable care to ensure the chatbot’s information was accurate.

The amount was modest, but the principle wasn’t. Companies are publishing more artificial intelligence-generated content than ever, including web copy, blog posts, comparison pages and sales collateral, and when that content is wrong, “the AI did it” doesn’t hold up as a defense.

Blaming a vendor doesn’t work either. Sports Illustrated published product reviews under fake author names with AI-generated headshots. Publisher The Arena Group blamed a third-party content vendor, AdVon Commerce, and said the articles were written by humans using pen names. The articles still ran under the Sports Illustrated name, and Arena Group’s shares fell more than 22% after the report. Earlier that year, CNET issued corrections to more than half of its AI-generated articles after finding factual errors.

The same risk applies across all industries, where marketing teams are under pressure to publish at volume. AI writing tools can turn out a competitor comparison page or an “industry analysis” in minutes. When those pages include claims nobody checked, like a competitor’s legal troubles, a customer’s supposed departure, or a capability that doesn’t exist, the company that published them owns the result.

The damage from that kind of content doesn’t stay on one page. Once published, AI-generated claims can be picked up and repeated by the AI tools that buyers use for research.

Wolf River Electric learned how that plays out. The Minnesota solar installer sued Google after its AI Overview search feature told users the company was being sued by the state attorney general. According to the complaint, none of the sources the AI Overview cited supported that claim. Wolf River is seeking $110 million to $210 million in damages and described a customer who raised concerns about the supposed lawsuit and terminated a $150,000 contract.

Google may have a defense that an ordinary publisher doesn’t. The company moved the case to federal court, and University of Minnesota Law School Dean William McGeveran said federal courts have been receptive to Section 230 defenses. “The question in this case is, ‘Who wrote the AI Overview?'” McGeveran said. For a vendor that publishes AI-generated copy on its own website, the answer is already clear.

This matters more now because of how buyers shop. A March, 2026 G2 survey of 1,076 B2B software buyers and decision-makers found that 51% now begin their purchasing process in an AI chatbot rather than a traditional search engine. The same report found that 85% of buyers think more highly of a vendor when an AI chatbot mentions them in a recommendation. The reverse is also true: A false negative claim that gets into those answers can reach every buyer who asks.

Hedged language doesn’t offer much protection, either. Content that says a company “may be facing” problems or “reportedly” lost business can look defensible to the person who wrote it. Research suggests AI tools often drop those qualifiers. A study published in Royal Society Open Science in April, 2025 compared 4,900 AI-generated summaries to their original texts, and found that most models produced broader generalizations than the source, even when explicitly prompted for accuracy, with newer models tending to perform worse. The study looked at scientific papers, but the mechanism is the same. “May be facing litigation” can easily become “is facing litigation” by the time it reaches a buyer.

Trax Technologies, a freight audit and payment provider, has filed a complaint against Freehand Inc. in the U.S. District Court for the Northern District of California. The complaint alleges false advertising under the Lanham Act and trade libel. It claims that statements about Trax’s Prizma.AI platform in Freehand’s marketing materials, including a comparison page, ranked comparison articles and case studies, were false or misleading. Trax seeks injunctive relief, damages and disgorgement of profits. 

For companies publishing AI-assisted content, the fix is mostly process. Every factual claim needs a source. Claims about other companies need extra review. Someone with a name signs off before anything goes live, whether the draft came from an employee, an agency or a model. When something wrong gets published, it’s an immediate liability, because even unpublishing doesn’t remove it from the internet archives. What’s more, AI tools keep reading the original page.

In supply chain operations, audits exist because AI summaries of vast data sets aren’t bulletproof evidence. The same principle applies to everything a company publishes under its own name.

Blake Tablak is chief executive officer of Trax Technologies.