Your Chatbot Said It, So You Own It: What the Air Canada Ruling Means for Small Businesses

In February 2024, a small tribunal in British Columbia decided a dispute worth a few hundred dollars. It is still one of the most quoted stories in the chatbot world, because it answered a question many business owners had been quietly avoiding: if your website chatbot gets something wrong, whose problem is it?

1. What Actually Happened

According to McCarthy Tétrault's summary of the case, a customer named Jake Moffatt asked Air Canada's website chatbot about bereavement fares and was told he could apply for the discount after travelling. Another page on the airline's own site said something different. When he later claimed the discount, it was refused, and he took the matter to the British Columbia Civil Resolution Tribunal.

The tribunal, in a decision issued on 14 February 2024 (2024 BCCRT 149), found the airline liable for negligent misrepresentation. Reported damages came to $650.88, with interest and tribunal fees bringing the total to $812.02. The amounts are small. The reasoning is what people remember.

The detail that makes this case instructive is how ordinary it was. This was not a sophisticated attack or an obscure edge case. A customer asked a routine question about a policy, got a clear answer in writing, and acted on it. Any business with a website chatbot could face the same situation tomorrow.

2. The Argument That Failed

Air Canada argued, in effect, that the chatbot was a separate thing and the airline could not be held to what it said. The tribunal rejected that. As reported, it treated the chatbot as part of the airline's website and held the company responsible for all information on that website, whether it came from a static page or a chatbot.

It also noted that the customer had no reason to know which part of the site to trust. If two pages disagree, a visitor should not have to guess which one is right.

Consider what the opposite outcome would have meant. If a business could disown anything its chatbot said, customers would have no reason to trust any chat answer, and businesses would have a free way to avoid obligations. Seen that way, the tribunal's reasoning is not surprising. A customer deals with "the website", and the website includes the chat window.

3. What the Ruling Does Not Prove

It is worth being precise here, because this case gets overstated. It was decided by a small-claims tribunal, and commentators note that such a decision is not binding precedent for courts. The facts were specific: a clear, wrong statement, a customer who relied on it, and a loss. Other courts in other places may reason differently.

• It does not mean every chatbot error leads to a payout. It means a clearly wrong answer that a customer reasonably relied on can create real liability.
• It does not create a universal rule. Check how your own jurisdiction treats website statements, and ask a qualified professional if the stakes are high.
• It is not legal advice from us. We are describing a public case, not advising on your situation.

The honest takeaway is a middle path. Do not panic, and do not dismiss it. The case shows how a decision-maker might reason about chatbot statements, and it is a good prompt to check your own setup before a customer does it for you.

4. Why Plain AI Chatbots Are Exposed

A chatbot that answers from general training rather than your content can produce a fluent, confident, invented policy. We cover the mechanism in why bots make up answers. The Air Canada case is what that failure looks like when a real customer acts on it.

The risk is highest for the questions where a wrong answer costs money or trust: refunds, cancellations, warranties, prices, delivery dates and eligibility rules.

Think about the questions where a wrong answer is costly: "Can I get a refund after the trip?", "Is this covered by warranty?", "Will it arrive before Friday?". A model answering from general training will produce something plausible for each, in the same confident tone it uses for trivia. Nothing in the answer tells the visitor, or you, that it was a guess.

That is why the safer design is to give the bot a limited job: answer from your approved content, and say so plainly when that content does not cover the question.

5. A Practical Checklist for Your Own Site

You cannot control what a customer asks, but you can control what your bot is allowed to say. These five steps reduce the chance of a costly wrong answer.

1. 1Make your policies the single source of truth. If your site contradicts itself, the chatbot cannot fix that. Resolve conflicts between pages first.
2. 2Restrict the bot to approved content. A grounded chatbot answers only from the material you give it, and says so when it has nothing.
3. 3Give money questions extra care. For refunds, cancellations and pricing, write the answer yourself and keep it current.
4. 4Always offer a human route. When the bot is unsure, it should hand over to a person rather than improvise. See why "I don't know" is the best answer.
5. 5Review what it says. Read real conversations regularly and fix the answers that drift.

If you only have an hour, do these in order: fix any contradictions between your own pages, write explicit answers for your five riskiest topics (refunds, cancellations, pricing, delivery, eligibility), and turn on a visible route to a human. Those three changes remove most of the exposure the Air Canada case describes.

Then make review a habit. Read a sample of real chats every week for the first month and monthly after that. You will spot drift, confusing wording and gaps in your content long before they turn into a customer dispute.

6. Questions to Ask Your Chatbot Vendor After Reading This

The ruling is a useful prompt for a conversation with whoever supplies your chatbot. You are not trying to catch them out; you want to understand how the product protects you and what remains your responsibility.

• Can I restrict the bot to content I approve? Ask to see how, and to test a question outside that content.
• What does it say when it does not know? Ask for a live demonstration, not a description.
• Can I read the conversation history? You need to be able to see what it told customers.
• How quickly can I correct a wrong answer? A fast fix reduces the time a mistake can affect customers.
• Does it provide a clear route to a person? A handoff protects both the visitor and you.

Keep in mind that no vendor can take over your legal responsibility for what appears on your website. What a good vendor can do is give you the controls and visibility to manage that responsibility sensibly.

If your business handles high-stakes topics such as health, finance or legal matters, consider adding extra review steps and consult a qualified professional about your specific obligations. For routine questions, the practical steps in this post are a solid starting point.

Summary

The Air Canada ruling is small in money and large in message: customers do not distinguish between your static pages and your chatbot, and neither should your risk planning. You cannot remove the risk by calling the bot a separate tool. You reduce it by controlling what the bot is allowed to say and by giving it a safe way to hand over.