Your Chatbot's Promises Are Legally Binding. Air Canada Found Out the Hard Way.

In plain terms: If you put an AI on your website or your phone line, whatever it tells a customer counts as something your business said. A court has already ruled on this. "The computer made it up" is not a defense.

A man named Jake Moffatt went to Air Canada's website after his grandmother died. The chatbot told him he could book a flight now and apply for the bereavement discount retroactively, within 90 days.

That policy did not exist. The chatbot invented it.

Air Canada refused the refund. Moffatt took it to the British Columbia Civil Resolution Tribunal, and in February 2024 the tribunal ruled against the airline. The part worth your attention is not the money. It is the argument Air Canada tried.

The defense that failed

Air Canada argued that the chatbot was, in effect, a separate entity responsible for its own statements.

The tribunal rejected that outright. The chatbot was part of Air Canada's website. Air Canada was responsible for all the information on its website, whether it came from a static page or a conversational bot. The airline was found liable for negligent misrepresentation and ordered to pay damages.

The dollar figure was small, a few hundred dollars. The precedent is not. A company does not get to distance itself from the thing it deployed to talk to customers.

Why this matters more for you than for Air Canada

An airline can absorb a bad ruling. It has a legal department, a policy team, and a brand large enough to survive a news cycle.

You are a shop with a phone that rings. If an AI on your line tells a customer you offer a warranty you do not offer, quotes a price you cannot honor, or promises a Saturday appointment you do not work, that is your problem on Monday. Not the vendor's. Not the model's.

And here is the uncomfortable part: the failure mode is not the AI going haywire. It is the AI being helpful. These systems are built to give a satisfying answer. Asked a question slightly outside what they know, the pull is toward inventing something reasonable rather than admitting the gap. A confident wrong answer feels like good service right up until it is a liability.

So the real work is not the bot

Everybody selling you an AI receptionist is selling the conversation. That part is close to a commodity now. The models are good and getting better, and the demo always sounds great, because a demo is a conversation with nothing at stake.

The part that actually protects you is boring and nobody demos it:

What it may state as fact. Your hours, your service area, whether you are taking new customers. Short, true, and written down.

What it may never state. Pricing beyond a published range. Timelines. Warranty terms. Anything about a specific job it cannot see. Guarantees of any kind.

Where it hands off. Every question outside the allowed set has to have a destination: a text to you, a callback promise it can actually keep, a human. A bot with no escape hatch will improvise, because improvising is what it does.

What gets logged. Every conversation, retrievable. When a customer says "your system told me," you need to be able to open the transcript and find out whether they are right. Sometimes they will be. That is worth knowing before they tell you in a review.

Those four things are configuration and operating procedure. They are not a prompt, and you cannot buy them off a shelf, because they are specific to what your business actually promises.

The honest test

Before you put an AI in front of a customer, ask the person selling it one question: what happens when it says something we do not honor?

If the answer is a shrug, or a claim it will not happen, you have your answer about who is carrying that risk.

If the answer is a documented scope, an escalation path, and a log you can read, you are talking to somebody who has thought about the Monday after.

Air Canada's chatbot did not have a bad day. It had a bad boundary. Somebody had to decide what it was allowed to promise, and nobody did.

That decision is the deliverable. The conversation is just the part you can hear.

---

If you want the version where every call gets answered and nothing gets promised that you cannot honor, that is Speed to Lead. The scope limits come before the bot, not after.