Is your business responsible when its chatbot gives a customer a wrong answer?

A customer asks your website's chatbot whether a sale item can be returned. The chatbot, built on a language model, answers confidently: "Yes, within 60 days." Your actual policy is 14 days, final sale on clearance. Who is bound by the answer? For practical purposes, assume you are. A chatbot is not a separate person that can be blamed; it is a feature of your website, like a price tag or a policy page, and customers are entitled to rely on what your business tells them.

That principle is not new, but generative AI makes it urgent. A language model writes fluent answers even when it is wrong, an error often called a hallucination. A scripted chatbot could only repeat what you typed into it. A generative one can invent a refund rule, a delivery date or a discount. This article looks at what the law says as of October 2026 and how a small business can set up a chatbot that stays within what it is allowed to promise. It is educational, not legal advice for your specific case.

What the law says in Québec

Québec's Consumer Protection Act was written long before chatbots, and it does not mention them. Several of its rules are written broadly enough that they don't need to:

How a Québec court would apply these sections to a particular chatbot is for the court to decide, but nothing in them suggests that a merchant can escape responsibility because software wrote the sentence.

Other places are making that explicit. Utah's 2025 law on generative AI in consumer transactions says it is not a defense to a consumer protection violation that generative AI made the statement or did the act. Expect more laws to follow that pattern.

A worked example: a bike shop's chatbot

A bike shop in Gatineau adds an AI chatbot to its website to answer questions in the evening. Before launch, the owner lists the topics where a wrong answer would cost money or trust, and decides what the chatbot may do with each:

The owner then writes the chatbot's system prompt, the standing instructions it reads before every conversation:

You are the website assistant for a bicycle shop. Answer only from the shop information, product catalog and policy documents provided to you. Never invent prices, discounts, delivery dates or policy exceptions. If a customer asks for a refund, return or warranty decision, quote the relevant policy, link to it, and offer to send the question to a staff member. If you are not sure, say so and offer the staff contact form. Always mention conditions such as "final sale" or delivery fees when they apply.

A system prompt reduces errors but does not eliminate them, so the owner also tests it: thirty tricky questions, run before launch and again after every change to the policies.

What does the Air Canada chatbot case teach small businesses?

In November 2022, the day their grandmother died, Jake Moffatt asked the chatbot on Air Canada's website about bereavement fares. It said the customer could buy tickets at the regular price and apply for the reduced fare afterwards, within 90 days. Air Canada's own bereavement page said the opposite: no requests after travel. When Moffatt applied, the airline refused, although an employee later admitted the chatbot had used misleading words.

Before British Columbia's Civil Resolution Tribunal, Air Canada argued it could not be held liable for what its chatbot said. The tribunal called that a remarkable submission, held that the chatbot was simply part of the airline's website, and found the airline responsible for all the information on its site. It found that Air Canada had not taken reasonable care to make the chatbot accurate, that relying on it was reasonable, and that customers should not have to double-check one part of a website against another. The airline was ordered in February 2024 to pay $812.02 in damages, interest and fees.

It was a small-claims decision in another province, and it does not bind Québec courts. Its lessons are still practical:

• Your chatbot speaks for you. Plan as if every answer were written by a staff member.
• One source of truth. If the chatbot and your policy page disagree, the customer is not the one who has to sort it out. Connect the chatbot to the same documents your staff use.
• Fix fast and keep records. When a wrong answer surfaces, correct the chatbot, honor reasonable reliance where you can, and keep the conversation logs.
• Small amounts still count. A few hundred dollars per mistake adds up across many conversations, and so does the damage to trust.