In November 2022, Jake Moffatt's grandmother died in Ontario. That same day, Moffatt went to Air Canada's website to book a flight from Vancouver to Toronto and asked the site's chatbot about bereavement fares. The chatbot said that anyone who had already travelled could submit their ticket for a reduced bereavement rate within 90 days of the date it was issued.
That wasn't Air Canada's policy. The airline's own 'Bereavement travel' page, which the chatbot linked to in the same answer, said the policy didn't apply to requests made after travel was completed. Relying on the chatbot, Moffatt bought two regular-fare one-way tickets, which the tribunal totaled at $1,630.36, applied for the partial refund well inside 90 days, and was turned down.
What the tribunal decided
| Date | What happened |
|---|---|
| Nov. 11, 2022 | Moffatt asks the chatbot about bereavement fares and books an outbound flight relying on its answer |
| Nov. 17, 2022 | Moffatt submits the first request for the bereavement rate, within the chatbot's 90-day window |
| Feb. 8, 2023 | An Air Canada representative concedes the chatbot used 'misleading words' and says it will be updated |
| Feb. 14, 2024 | British Columbia's Civil Resolution Tribunal finds negligent misrepresentation and orders Air Canada to pay $812.02 |
Air Canada argued it couldn't be held liable for information from its agents, servants or representatives, including a chatbot. Tribunal member Christopher C. Rivers called that 'a remarkable submission.' A chatbot may be interactive, he wrote, but 'it is still just a part of Air Canada's website,' and the airline is responsible for all the information on its website, whether it comes from a static page or a chatbot.
He also rejected the idea that the customer should have checked the policy page. Air Canada didn't explain why that page was more trustworthy than its chatbot, or 'why customers should have to double-check information found in one part of its website on another part of its website.'
The award covered the fare difference ($650.88), pre-judgment interest ($36.14) and tribunal fees ($125). It's a small-claims decision from a Canadian tribunal and doesn't bind US courts, and this post is general information rather than legal advice; rules vary by state and province.
Still, the reasoning is plain enough that 'the bot said it, not us' is a hard argument to make anywhere. A similar complaint in the US might be framed as misrepresentation or under a state consumer-protection law, and how that plays out depends on your state, which is one more reason to bring in a lawyer early. In the US, when the Federal Trade Commission announced a 2024 crackdown on deceptive AI claims, its chair said 'there is no AI exemption from the laws on the books.'
Three lessons for any business with a chatbot
- 01The chatbot speaks for you. Customers can't tell which part of your website is the official one, and the tribunal said they shouldn't have to.
- 02Two sources of truth is one too many. The chatbot contradicted a policy page on the same site, in an answer that linked to that page. If your chatbot paraphrases policies from its own instructions or from old documents, sooner or later it will disagree with the page.
- 03Problems surface late unless you look. Air Canada's representative acknowledged the error in February 2023, a year before the decision. Reading transcripts weekly against the policy would have caught it before anyone relied on it.
Where a small business chatbot makes the same mistake
Most small businesses don't sell bereavement fares. They have the equivalent: deposit terms, insurance coverage and prices that sound like promises when a chatbot says them.
| Business | Risky answer | Safer answer |
|---|---|---|
| Restaurant | 'Yes, the private room deposit is refundable if you cancel.' | 'Here's our events cancellation policy. The events manager confirms terms for your date.' |
| Dental practice | 'Cleanings are fully covered under your plan.' | 'According to the insurance list we keep current, we're in network with that plan. We'll check your benefits before the visit and give you an estimate.' |
| Carpenter or remodeler | 'A 12-by-16 deck runs about $8,000.' | 'Decks that size usually fall within the range on our pricing page. We set the price after a site visit.' |
| Any business | 'We'll match that competitor's price.' | 'I'll pass that to the owner, who will reply today.' |
Flowchart
How a chatbot should handle a question about money or eligibility
- StartCustomer asks: 'Can I get my deposit back if we cancel the party?'
- StepAssistant finds the approved events policy page
- DecisionWould answering commit us on price, a refund or eligibility for this customer?YesA personAssistant quotes the policy, then hands off; the events manager confirms this customer's termsNo
- StepAssistant quotes the policy word for word and links to the page
- StepConversation logged along with the page it quoted
- DoneWeekly review checks that answers still match the page
Before and after
How the Air Canada setup failed, and the fix
Two sources of truth
- 01Policy lives on a webpage
- 02Chatbot produces its own summary of the policy
- 03The summary drifts away from the page
- 04Customer relies on the chatbot and buys
- 05Business learns of the error from a complaint
ResultA promise the business never meant to make
One source, with handoffs
- 01Policy lives on one approved page
- 02Chatbot quotes that page word for word, with a link
- 03Policy changes update the page, and the chatbot follows
- 04Refund and eligibility requests go to a person
- 05Transcripts reviewed weekly against the page
ResultAnswers that match the policy, or reach a human
A 30-minute audit of the chatbot you already have
If a chatbot is already live on your site, you can find the Air Canada problem in half an hour with a notepad and your policy pages open in another tab.
- 01List the five policies customers ask about most: deposits, cancellations, insurance or warranty coverage, pricing and scheduling.
- 02Ask the chatbot about each one three ways, including a leading version such as 'So I can get a full refund, right?'
- 03Compare every answer with the policy page, line by line.
- 04Mark any answer that adds a condition, drops one or makes a promise the page doesn't.
- 05For each mark, fix the source the chatbot used or add the topic to the handoff list, then ask again.
- 06Finally, check that the chatbot says it's automated and offers a way to reach a person.
The leading questions matter most. Many chatbots are built to be agreeable, and 'right?' invites a yes.
Guardrails to set before launch, or this week
- Approved content only. The chatbot answers from pages and documents you've signed off, not from the model's general knowledge.
- Quote policies, don't paraphrase them. Refunds, cancellations, warranties and coverage should be quoted with a link.
- A handoff word list. Refund, deposit back, guarantee, covered, free, discount, price match, warranty, deadline, 'can you promise.'
- A visible path to a person. A phone number, callback request or form in every conversation.
- Disclosure. Say at the start that it's an automated assistant.
- Logs that someone reads. Keep transcripts with the source each answer used, and read a sample every week.
- A plan for errors. When a wrong answer turns up, fix the source, check whether anyone relied on it and decide with the owner how to make it right.
If a customer has already relied on a wrong answer, talk to your lawyer before you reply. Whether to honor it is a business and legal judgment, not a chatbot setting. For the content side, our guide to preventing made-up answers explains how to test a chatbot before customers do, and we build website chatbots with these limits from the start.