Who Is Liable When Your Chatbot Gets It Wrong?
In Moffatt v. Air Canada, a BC tribunal held the airline liable for its chatbot's wrong answer. What it means and how to set up an assistant that's far less likely to guess.
Assume you are. In Moffatt v. Air Canada (2024), a British Columbia tribunal held the airline responsible for a wrong answer its website chatbot gave a customer. The practical lesson for a small business: an assistant should answer only from your information, say when it doesn’t know, and pass anything involving money or promises to a person.
The Civil Resolution Tribunal’s decision ordered the airline to pay $812.02 in damages, interest and fees.
This article is general information, not legal advice. For your own situation, speak to a lawyer.
What you want is an assistant that takes repeated questions off your plate without ever putting words in your mouth. The case shows what happens when that goes wrong, and the rest of this article shows how to set one up to make that much less likely.
What happened in Moffatt v. Air Canada?
In November 2022, after their grandmother died, Jake Moffatt used the chatbot on Air Canada’s website while booking a flight from Vancouver to Toronto (2024 BCCRT 149). According to the decision, “The chatbot suggested Mr. Moffatt could apply for bereavement fares retroactively” (Civil Resolution Tribunal, 2024 BCCRT 149). Air Canada’s own bereavement page said otherwise, and the airline refused the partial refund (2024 BCCRT 149).
The tribunal, in a decision issued on 14 February 2024, 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 (decision).
Why did the tribunal reject the “the chatbot did it” defence?
Because the chatbot is part of the website. The tribunal wrote that Air Canada, in effect, “suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission.” It continued: “It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot” (Civil Resolution Tribunal, 2024 BCCRT 149).
It also found that Air Canada “did not take reasonable care to ensure its chatbot was accurate”, and that the airline did not explain “why customers should have to double-check information found in one part of its website on another part of its website” (2024 BCCRT 149). The chatbot’s reply had included a link to the correct policy page (2024 BCCRT 149). That wasn’t enough.
Does this apply to a small business in Ontario?
The decision comes from BC’s small-claims tribunal, not from an Ontario court, and we’re not lawyers. It’s also a decision on its own facts: Air Canada didn’t file the tariff terms it relied on or evidence of the actual bereavement fare, and the tribunal noted both (2024 BCCRT 149). But the tribunal’s reasoning is plain enough to act on: a business answers for what its website tells customers, and a chatbot is part of the website. If your chatbot tells someone they’ll get a refund, a discount or a same-day appointment, assume a customer can reasonably rely on it. Ask a lawyer how this applies to your business and your policies.
Which chatbot answers carry the most risk?
The riskiest answers are the ones that commit you to something. Here’s how they compare:
| Type of question | Example | Risk if the answer is wrong | Safer design |
|---|---|---|---|
| Facts you’ve published | ”What time do you close Saturday?” | Low, if your information is current | Answer from your hours |
| Money and refunds | ”Can I get my deposit back?” | High: a promise about money | Quote your written policy word for word, or hand off |
| Exceptions | ”Can you make an exception for a bereavement?” | High: close to Moffatt, where the chatbot misstated a bereavement fare policy | Always hand off to a person |
| Availability | ”Can you come tomorrow?” | Medium: a missed booking | Hand off, or link to your booking tool |
| Professional advice | ”Is this rash serious?” | High: outside any chatbot’s role | Refuse and point to a professional |
| Anything not in your information | ”Do you price-match?” | High if it guesses | Say “I don’t know” and hand off |
How do you set up an assistant that’s less likely to guess?
Five rules, which work with any provider:
- Answer only from information you’ve approved. Your services, hours, service area and written policies. Not the AI model’s general knowledge.
- Say “I don’t know” out loud. When the answer isn’t in your information, the assistant should say so and offer to pass the question on.
- Hand off money and exceptions. Refunds, discounts, deposits and “can you make an exception” go to a person, every time.
- Keep the source current. An assistant reading last year’s policy gives last year’s answer. Update its information the same day you update your website.
- Show where the answer came from. An assistant that can point to the page it used makes errors easier to spot.
We cover how this works from end to end, including what to give the assistant, in AI assistants for small business.
How do you test for risky answers before launch?
Ask it the questions from the table above, in your customers’ words, before anyone else can. Include at least three it should refuse: a refund request, an exception, and something your information doesn’t cover. Any answer that sounds like a promise you haven’t written down is a fail. Our AI assistant guide describes the assistant’s three-question test, which you can repeat whenever your information changes. For the website around it, use the 2-minute checkup in our small business website guide for Waterloo Region.
What should you ask a chatbot vendor?
“Where do its answers come from?”, “What does it do when it doesn’t know?”, “How does a customer reach a person?” and “How will we test it before launch?”. If the vendor says it “can answer anything”, keep looking. These questions are part of our list of questions to ask before hiring a web designer.
A hypothetical example
Picture a dental office in Waterloo whose assistant is asked, late on a Sunday, “If I cancel tomorrow’s appointment, will I be charged?” The office’s written policy has a cancellation fee, with exceptions decided case by case. The obstacle is the exception: an assistant that guesses might say “no charge”. One limited to the office’s information quotes the written policy and passes the request for an exception to the front desk with the patient’s contact details. It’s an illustration, not a case study, and not legal advice.
Next step
Next read: AI assistants for small business, for the full picture of how a well-behaved assistant works, and Canadian cases where AI got it wrong, for what courts said after Air Canada.
If you already have a chatbot, run the risky questions from the table against it today. If you’re choosing one, see the five steps and their prices at /express. The Express assistant is set up to answer from your own information, point to where an answer came from, say “I don’t know” when the answer isn’t there, and pass the question and the customer’s contact to you. No assistant is error-free, and you remain responsible for what it tells customers, so ask it your own risky questions before it goes live, whoever builds it. You send your policies and FAQ; you keep your booking tool. Each step shows a range in business days, and the clock starts when we have your content. Name and email are enough to get the price in writing. Local details are in our AI chatbot guide for small businesses in Waterloo.
Sources checked on 29 September 2026. This article is general information, not legal advice. Examples are hypothetical.
How we wrote this: drafted by the YourFirstAI team with AI assistance, sources checked on 2026-09-29, and reviewed against our editorial rubric before publishing. Examples are hypothetical unless they cite a source.
Frequently asked
Did Air Canada argue its chatbot was responsible for itself?
Not in those words. Air Canada argued it couldn't be held liable for information given by its agents or representatives, including a chatbot. The tribunal said that, in effect, this suggested the chatbot was a separate legal entity responsible for its own actions, called it a remarkable submission, and found the airline responsible for all the information on its website.
How much did Air Canada have to pay?
The tribunal ordered Air Canada to pay $812.02: $650.88 in damages, $36.14 in pre-judgment interest and $125 in tribunal fees. It dismissed the customer's remaining claims.
Does a link to my policy page protect me if the chatbot is wrong?
It didn't in this case. The chatbot's answer linked to Air Canada's bereavement page, but the tribunal said the airline didn't explain why customers should have to double-check one part of its website against another. Get legal advice for your own situation.
Is this decision binding in Ontario?
It's a small-claims decision from British Columbia's Civil Resolution Tribunal, not a ruling from an Ontario court. How it would apply to your business is a question for a lawyer. This article is general information, not legal advice.
What's the simplest way to reduce the risk?
Limit the assistant to your own approved information, have it say it doesn't know when the answer isn't there, and pass refunds, prices and exceptions to a person. Test it with tricky questions before customers see it.