When your brand's bot answers a customer, you answered

In February 2024, a Canadian tribunal ruled on a small case. It matters to every marketer running AI on customer channels.

Jake Moffatt asked Air Canada’s website chatbot about bereavement fares. The bot told him he could claim the discount afterwards. Another page on the airline’s own site said the opposite. Air Canada refused the refund. Then it argued the chatbot was a separate entity responsible for its own actions.

In Moffatt v. Air Canada (2024 BCCRT 149), the tribunal found the airline hadn’t taken reasonable care to keep its chatbot accurate. It held the airline liable for negligent misrepresentation and called the chatbot simply part of the website. According to the McCarthy Tétrault summary I relied on (I couldn’t open the decision itself), it said it makes no difference whether information comes from a static page or a chatbot.

A small-claims tribunal binds no other court. But I’d expect most regulators and judges to find the logic convincing. If your brand’s account says something, your brand said it.

Customers want speed, and the law wants honesty

Sprout Social reports that:

Its page gives no sample sizes, so treat these as rough signals. The same goes for its finding that 47% of marketers plan to use AI for basic inquiries. The pressure to automate makes sense. The question is what to automate.

Europe answers part of it. From 2 August 2026, Article 50(1) of the AI Act requires providers to build AI systems that talk directly to people so those people know they’re dealing with an AI. The exception is when that’s obvious to a reasonably observant person.

The Digital Omnibus delayed the high-risk rules to December 2027 and August 2028. But according to a Jones Walker analysis, it left most transparency duties in place. Breaches sit in the tier that carries fines of up to €15 million or 3% of worldwide turnover.

The design duty falls on providers. But the brand that chose the tool is the one facing the customer. Article 50(4) also asks deployers to disclose realistic AI-made or altered images, audio and video.

I’m not a lawyer, and the EU text is the authority. My practical reading: if a bot answers customers, say so.

A reply ladder

Here’s how I’d design a social care operation. It’s my opinion, informed by the sources here.

AI answers alone only for acknowledgements and for facts that live in one approved source. Think opening hours, delivery windows, order status from the live system. If the answer isn’t there, the bot says it doesn’t know and hands over.

AI drafts and a person sends for product questions, how-to help and anything with nuance. The draft shows which source it used.

People only for complaints, refunds and compensation, safety, legal matters, the press, vulnerable customers and anyone angry. Each category gets a named owner and a response time.

Air Canada failed on the first rung. The bot contradicted the airline’s own policy page, and nothing made it check. One source of truth, plus a rule that the bot answers only from it, would have prevented it.

What does accountability look like?

Keep one maintained set of policies and facts for the bot, with a named owner. If two pages disagree, the bot will find the wrong one.

Tell people at the start that they’re talking to a bot. Log every automated answer with its source. Have someone score a random sample each week for accuracy and tone. Make sure one person can switch the bot off within a minute.

Then measure what the customer cares about. Was the problem fixed, and how fast?

A quick acknowledgement that leaves the problem untouched makes a brand look responsive and incompetent at once. A bot that closes more tickets because customers gave up is worse than no bot. So pair deflection with satisfaction on the tickets it handled, and the share that reopened.

Write the reply ladder before buying the bot. A working version of the triage workflow, with its checks, is in my AI workflows for social media teams. The same thinking applies to what you publish.

Sources