The chatbot that invented a refund policy — and the airline had to honour it

On 11 November 2022, the day their grandmother died in Ontario, Jake Moffatt went to Air Canada's website to book a flight from Vancouver to Toronto, and asked the support chatbot on the site about the airline's bereavement fares [1].

The chatbot answered that if you need to travel immediately or have already travelled, you could submit the ticket for a reduced bereavement rate "within 90 days of the date your ticket was issued". Air Canada's own bereavement-travel page — linked from inside that very answer — said the opposite: the policy does not apply to requests made after travel is complete [1].

Moffatt booked the flights, paying $1,630.36 in all, and applied for the reduction on 17 November, well inside the window the chatbot had described. Air Canada refused. In February 2023 one of its representatives admitted the chatbot had used "misleading words" and said the company had noted the issue so it could update it. The parties could not resolve matters, and Moffatt brought the claim to British Columbia's Civil Resolution Tribunal [1].

Air Canada argued it could not be held liable for information given by its own chatbot — suggesting, in effect, that the chatbot was a separate legal entity responsible for its own actions. Tribunal member Christopher C. Rivers called that "a remarkable submission" and rejected it: a chatbot is still just part of Air Canada's website, and it makes no difference whether information comes from a static page or a chatbot [1][2].

The tribunal was equally unimpressed by the argument that Moffatt should have cross-checked the bot against another page, noting that Air Canada never explained why one part of its website was inherently more trustworthy than another, or why customers should have to double-check it against itself [1].

On 14 February 2024 the tribunal found negligent misrepresentation and ordered Air Canada to pay a total of $812.02: $650.88 in damages, $36.14 in pre-judgment interest, and $125 in tribunal fees [1][2].

Two weeks after the ruling, Klarna announced that its OpenAI-powered assistant had handled 2.3 million conversations in its first month — two-thirds of its customer-service chats, which it described as the equivalent work of 700 full-time agents, with satisfaction on par with human staff [3]. Notice how differently the two claims on this page are established: one is a tribunal's finding on the public record, the other a company's own figure about its own product. That distinction earned its keep. By May 2025 Klarna was recruiting human agents again, its chief executive saying cost had become too dominant a factor in how support was organised, and that the result was lower quality [4]. The technology worked. The claim was still worth checking.

The difference was never the model — it was what the assistant was allowed to do when it did not know. Air Canada's bot produced a fluent, specific, plausible policy: a real-sounding window, a real-sounding form, a real-sounding process. Nothing about the answer looked like a guess, because this kind of wrong answer never does [1]. An assistant that can only quote the published policy and link to it cannot invent a 90-day window. One that writes prose can, and did.