The short answer: yes, a business is responsible for what its AI agent tells a customer, in the same way it is responsible for what its written pages say. In February 2024, a Canadian tribunal ordered Air Canada to pay a passenger after its website chatbot invented a refund policy that did not exist. Air Canada argued the chatbot was to blame, not the airline. The tribunal rejected that outright. We build AI agents for a living, so the ruling is a useful, real test case for what happens when a chat agent is allowed to answer from anywhere rather than from a business's actual policies.
Key takeaways
- Air Canada's chatbot told a grieving customer he could claim a bereavement fare refund after booking his flights. The airline's real policy required the claim to be made before travel.
- The B.C. Civil Resolution Tribunal ruled in February 2024 that Air Canada was liable for negligent misrepresentation, ordering it to pay CAD $812.02, reported by the BBC at the time as roughly £642.64.
- The tribunal's reasoning was blunt: "it makes no difference whether the information comes from a static page or a chatbot." Air Canada's attempt to argue the chatbot was a separate, self-responsible entity was called "a remarkable submission."
- Our AI agents answer only from a business's real services, prices and policies, and hand anything outside that to a person before it reaches a customer. A site chat agent is a fixed £499, quoted in writing before we build it.
What Air Canada's chatbot actually got wrong
In November 2022, Jake Moffatt's grandmother died. The same day, he went to Air Canada's website to book a flight from Vancouver to Toronto and asked the site's chatbot about bereavement fares. According to the tribunal's published decision, the chatbot told him he could book at the full fare and then submit a claim for the reduced bereavement rate within 90 days of the ticket being issued. Moffatt booked two full-price flights on that advice, paying $1,630.36 CAD in total.
Air Canada's actual policy said the opposite: bereavement fares had to be requested before travel, not claimed back afterwards. When Moffatt applied for the refund, the airline turned him down, later admitting the chatbot had given "misleading words," and offered him a $200 coupon as a goodwill gesture. He refused it and took the case to the tribunal instead.
Why "the chatbot did it, not us" did not work
Air Canada's defence was that the chatbot was, in the tribunal's words, "a separate legal entity that is responsible for its own actions." Tribunal member Christopher Rivers did not accept it. As CBC News reported, his ruling stated plainly: "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." He also rejected the airline's argument that Moffatt should have checked a different page instead of trusting the chatbot, finding no reason a customer should know which part of a company's own site to trust.
Air Canada was ordered to pay $812.02 CAD: the fare difference, interest and tribunal fees combined. The ruling was made by a Canadian tribunal, so it carries no legal weight in the UK. But the substance of what went wrong is not specific to Canada or to airlines. A chat tool told a customer something the business's own policy did not support, and nobody had checked the tool against that policy before it went live.
The actual lesson for a small business
The failure here was not that Air Canada's chatbot was malicious or unusually broken. It is that a chatbot allowed to answer freely will sometimes produce an answer that sounds confident and is wrong, a pattern often called an AI hallucination. The fix is not a better disclaimer buried in the terms and conditions. It is limiting what the tool is allowed to answer from in the first place.
There is a real difference between an AI agent that can say anything a general model might generate, and one that can only draw from a fixed, current set of facts about one business: its actual prices, its actual services, its actual opening hours and policies. The first kind can invent a refund window that never existed. The second kind cannot, because that information is not in what it was given to work from.
What we build differently
Every AI agent we build works from a business's real services, prices and hours, nothing invented and nothing assumed. You review what it knows before it goes live, so an out-of-date price or a policy that changed last month gets caught before a customer ever sees it. Anything a customer asks that falls outside what the agent has been given, an unusual request, an edge-case question, a complaint, gets handed to you rather than guessed at.
A site chat agent that answers questions on your website costs a fixed £499. Missed-call text-back, voice answering and custom automations are quoted in writing per job, because what they involve varies business to business.
If the phone is where you are actually losing customers
A chat agent answers the questions on your website. If missed phone calls are the bigger leak, particularly for trades and clinics where most enquiries still come by phone, our voice agent answers every call, quotes your real prices and books the slot, day or night. We wrote about what actually happens when a lead is called back fast if you want the detail on that side of it. Most clients start with one or the other, then add the second once the first is proven.
What it costs
| What you get | Price |
|---|---|
| Site chat agent | £499 (fixed) |
| Missed-call text-back | Quoted in writing |
| Voice answering | Quoted in writing |
| Custom automations | Quoted in writing |
Every other price we charge is on the pricing page, published in full, with nothing held back for a call.
Common questions
Will an AI agent ever just make something up for my customers?
Not from what it has been given to work with. It only answers from your real services, prices and policies, reviewed by you before it goes live, and anything outside that scope gets escalated to you rather than guessed at.
Can this happen to a UK business, not just a Canadian airline?
The ruling itself only binds Canada. But the underlying problem, a chat tool answering beyond what the business's actual policy allows, has nothing to do with which country's tribunal hears the complaint. It is a business risk everywhere a chat agent is deployed without being checked against real, current facts first.
What happens when a customer asks something the agent does not know?
It says so and hands the conversation to you, rather than improvising an answer to fill the gap. That is a deliberate limit, not a shortcoming.
Can I see how my current site handles this kind of thing before I add an agent?
Yes, for free. The 38-check scan covers the technical and structural basics of your website in an instant report, before you spend anything.
Want the version of this with your name on it?
Run the free 38-check scan on your current site, or ask for the free homepage mockup. Both free, neither with any obligation.