What the Air Canada chatbot ruling means for what AI says about your company

A customer asked Air Canada's chatbot about bereavement fares. The chatbot answered, linked to a policy page that said something different, and the airline later argued before a tribunal that it could not be held liable for what its chatbot had said. The tribunal ordered it to pay.

The decision is Moffatt v. Air Canada, 2024 BCCRT 149, from a tribunal in British Columbia, dated 14 February 2024. Air Canada is not a Miscited customer. We use the case because the decision is public and the facts are unusually clear. This post is not legal advice.

What the chatbot said

The decision quotes the chatbot's reply, including this passage:

"…have already travelled and would like to submit your ticket for a reduced bereavement rate, kindly do so within 90 days…"

In the same reply, the words bereavement fares linked to Air Canada's own page on bereavement travel. According to the tribunal, that page said the bereavement policy does not apply to requests made once travel has been completed (paras 16 to 17). The answer and the page it linked to disagreed.

The argument that did not work

Air Canada's position was that it could not be held liable for information provided by its agents, including a chatbot. The tribunal called this "a remarkable submission" (para 27). It ordered Air Canada to pay C$812.02 in total (para 44).

The sum is small, and this is one decision from one tribunal. It does not tell you how a court elsewhere would rule on different facts. What it does show is how the defence sounds when a company makes it out loud: the bot talks to our customers, and what it tells them is somebody else's problem. One tribunal was not persuaded. We would not plan on the next one seeing it differently.

Your own bot and someone else's assistant

The case was about a company's own chatbot. Buyers also read answers about companies from assistants those companies do not run: ChatGPT, Gemini, Perplexity and others. The two situations differ in almost every way that matters for a fix.

The right-hand column is the harder one. When an assistant you do not run states your old price, there is no bot of yours to fix and no quick legal remedy. There is still a buyer reading the answer and acting on it. The lever you have is the record the assistant reads, which is why the work starts with finding each wrong claim and tracing it to a page. What the courts have said so far about that column is in when AI says something false about a business.

The failure is a familiar one

Take the legal outcome away and the Moffatt answer is a specific, checkable defect: a confident claim attached to a source that says something else. Checking whether a claim is true and checking whether its cited page supports it are separate tests, and this answer fails both. We described that pair of checks in how to fix a wrong AI answer about your company.

It is also the kind of answer a mention count scores as a success. The company is named, the tone is helpful, and the link points at the company's own domain. We set out why that passes a visibility dashboard and fails an accuracy check in AI visibility and answer accuracy are different things.

One more lesson sits in the facts. The dispute was about a single conversation, preserved well enough to be quoted. Your own checks cannot work that way. One transcript shows that an answer is possible, not how often buyers get it: one wrong reply in two asks reads as 50%, with a 95% interval of 9% to 91%. We worked through how many runs a rate needs in how many prompts an AI visibility number actually needs.

What to do about it

• If you run a chatbot, list the policies it can talk about, with the date each one took effect, and test the bot against them whenever a policy or the bot changes. Treat a link to your own page as a claim to check, not as proof.
• For assistants you do not run, ask the questions your buyers ask, split the answers into claims and compare each claim with your dated facts. The full method is in how to check what ChatGPT, Gemini and Perplexity say about your company.
• Keep dated records of what was said, where and when, and of every correction you ask for. If a wrong answer ever becomes a dispute, the record is what you will have.

If you would rather start from evidence than from a method, the free answer audit samples buyer questions about your company, compares the answers with facts read from your own site and returns the wrong answers it can evidence, with the pages they cited.

Sources

1. Moffatt v. Air Canada, 2024 BCCRT 149, 14 February 2024, paras 15 to 17, 27 and 44. Full text on CanLII.