By The short answer: Treat every answer your website chatbot gives as something your business said. In 2024, British Columbia's Civil Resolution Tribunal ordered Air Canada to compensate a customer who relied on its chatbot's wrong answer about bereavement fares, and rejected the idea that the chatbot was responsible for its own words. You can still use one safely: limit what it answers to approved information, give it a clear route to a person, tell visitors they are talking to AI, and check its answers on a schedule.
What happened in Moffatt v. Air Canada
The decision is short and worth reading in full. In Moffatt v. Air Canada, 2024 BCCRT 149, a customer booked a flight after the death of his grandmother. While researching fares, he used the chatbot on Air Canada's website, which suggested he could apply for a bereavement fare after travelling. The airline's own bereavement travel page said otherwise, and Air Canada refused the retroactive request.
Air Canada argued it could not be held liable for information provided by its agents or representatives, including a chatbot. The tribunal member called that "a remarkable submission" and found the chatbot was still just part of the airline's website. The decision says it makes no difference whether information comes from a static page or a chatbot.
The tribunal also rejected the argument that the customer should have checked the correct page. It found customers should not have to double-check one part of a website against another. Air Canada was ordered to pay $812.02, made up of $650.88 in damages, interest and tribunal fees.
What the decision does and does not settle
This is one small claims decision from British Columbia. It does not bind courts in Alberta or elsewhere, and it turned on ordinary negligent misrepresentation rules, not on any law written for AI. That is the useful part for a business owner: no new rule was needed. The standard it applied is that a company must take reasonable care to make sure its representations are accurate and not misleading. A chatbot answering customers is making representations on your behalf.
This post is not legal advice. If your assistant answers questions about pricing, refunds, warranties, eligibility or anything regulated, have your lawyer look at what it says.
Where chat assistants go wrong
Most wrong answers come from a few predictable places.
- Open-ended knowledge. An assistant that draws on everything it can find, or on a general model's training, will fill gaps with plausible guesses. Policy details are exactly where plausible and correct part ways.
- Stale sources. The refund policy changed in March, but the document the assistant reads was last updated a year ago. Two pages on your site that disagree give the assistant two answers to choose from.
- No stopping point. The assistant has no instruction to say "I don't know" or to pass the question to a person, so it answers anyway.
- Implied authority. An assistant that says "you're booked" or "that will be refunded" when it has no ability to book or refund anything has made a promise your team now has to deal with.
Take a home renovation company whose assistant is asked, "Is the labour warranty transferable if I sell the house?" If the approved answer does not cover transfers, a well-built assistant says it cannot confirm that and offers to pass the question to the office. A poorly built one produces a confident yes or no. Only one of those answers is safe to put on your website.
The privacy side: tell people it is AI
Liability for wrong answers is one risk. The other is what happens to the information visitors type in.
Canada's federal, provincial and territorial privacy regulators, including Alberta's, issued joint principles for responsible, trustworthy and privacy-protective generative AI in December 2023. For public-facing tools, they say people interacting with the tool should be aware they are interacting with generative AI. They also call for collecting only the personal information you need and entering personal information into AI tools only where authorized.
The Office of the Privacy Commissioner's page on AI, privacy and your business adds being open about how information is used, making AI tools explainable to users and limiting the sharing of sensitive or confidential information.
For a website chat, that turns into plain decisions: a visible label saying the visitor is talking to an AI assistant, a short note on what happens to what they type, and a form that asks only for what your team needs to follow up.
A checklist for this week
Open your website chat and work through this list. If you do not have an assistant yet, use it as your specification.
- Ask it ten real questions. Pull them from your inbox or phone notes, including at least three about money: prices, refunds, deposits or warranties. Record every answer.
- Mark each answer as correct, incomplete or wrong. Anything wrong about money, eligibility or timing is the first fix.
- Write down what it is allowed to answer. Services, common questions, limits and escalation rules, in one document with a named owner.
- Fix conflicting pages. If two pages on your site disagree, the assistant is not the only problem.
- Add a refusal rule. When the approved information does not cover a question, the assistant says so and offers a person.
- Check the handoff. Submit a test enquiry through the chat and confirm it reaches the right person, and how long it takes to get a reply.
- Remove implied authority. The assistant should never say something is booked, approved or refunded unless it can actually do that and your team has agreed it should.
- Label it. Tell visitors it is an AI assistant and link to your privacy policy.
- Put a date on it. Repeat the ten-question test monthly, and every time a price or policy changes.
Where Lumin fits
We build website and voice assistants that answer from approved information and hand off to your team when a question needs a person.
- An approved knowledge set covering your services, common questions, limits and escalation rules, maintained by a responsible person on your team.
- A visitor-controlled route to leave details and reach your team, with status that separates a saved enquiry, a delivered notification and an actual staff response.
- Scoped connections to your tools, where linking to a booking page does not give the assistant authority to create or change appointments.
See AI assistants or book a discovery call.