In 2022, Jake Moffatt needed to fly to his grandmother’s funeral. He asked Air Canada’s website chatbot about bereavement fares, and it told him he could book now and apply for the discount within 90 days. He did exactly that, and Air Canada said no, because the real policy required the request before travel. The chatbot had simply invented a policy that didn’t exist.
Moffatt took them to a tribunal over a few hundred dollars, and Air Canada’s defense was that the chatbot was “a separate legal entity that is responsible for its own actions,” essentially arguing the airline shouldn’t have to answer for what its own AI told a customer.
The tribunal wasn’t buying it, ruled Air Canada responsible for everything on its own website, chatbot included, and made them pay up.
The more interesting question is why a chatbot was answering that in the first place. Bereavement policy is a fixed rule set: file before travel, within a window, with documentation, the same correct answer every time with no reasoning involved. A decision tree handles that without ever getting creative, which is the whole point, while a chatbot is built to sound helpful even when it doesn’t actually know the answer.
Next time someone suggests building a chatbot for something, it helps to ask whether people are navigating a known set of options or asking genuinely open-ended questions, since only one of those actually needs language rather than a lookup table.
If you want to see how that instinct holds up somewhere with real stakes, take the two-minute quiz.
