Air Canada Chatbot Case: Is Your Store Liable?

The Air Canada chatbot case shows that a business can be liable when a customer relies on wrong chatbot advice. Australian stores face a comparable risk under consumer law, although the Canadian ruling does not decide an Australian dispute. Treat replies about your products and policies as statements your store makes to shoppers.

What Happened in the Air Canada Chatbot Case

Air Canada’s chatbot told Jake Moffatt that a bereavement discount could be claimed after travel. Moffatt’s grandmother had died, and they relied on that answer when buying flights. The airline refused the refund because its actual policy excluded claims after travel.

In Moffatt v. Air Canada, 2024 BCCRT 149, British Columbia’s Civil Resolution Tribunal found negligent misrepresentation. It awarded C$650.88 for the fare difference and ordered C$812.02 in total, including interest and tribunal fees. These amounts are Canadian dollars.

The tribunal rejected the suggestion that the chatbot was a separate entity responsible for its own statements. It treated the chatbot as part of Air Canada’s website. A link to the correct policy did not excuse the contradictory reply or require Moffatt to double-check it.

Would the Same Apply in Australia

The Canadian decision does not bind Australian courts, but Australian Consumer Law creates a similar risk for misleading store replies. The ACCC’s guidance on false or misleading claims applies across business communications, including websites. Intent to mislead is not required.

Misleading or deceptive conduct is prohibited. False representations about a consumer’s rights or remedies can also breach the law. A chatbot that tells a shopper they have no refund rights can therefore create a consumer-law issue, even when the store’s written policy is correct.

The ACCC’s AI industry snapshot specifically warns about chatbots giving wrong information on consumer guarantee rights. It says businesses using chatbots need responses that comply with their consumer-law obligations. That supports treating your chatbot as a customer communication channel, without predicting the outcome of any particular claim.

There is also an Australian enforcement example involving a chat widget. In 2026, the Federal Court ordered JustAnswer to pay $10 million in penalties. The case covered misleading pricing statements made through its widget and misleading affiliation claims. It does not establish a rule specifically about generative AI.

Responsibility to the shopper and recovery from your chatbot vendor are separate questions. Your vendor contract can affect whether you recover losses, through indemnities, liability limits or other terms. It does not by itself settle who owes the shopper a remedy. Have a lawyer review that contract, your chatbot disclaimers and your policy wording. This page gives general information, not legal advice.

Where a Store Chatbot Can Mislead Shoppers

A risky reply can change a shopper’s buying decision or stop them exercising a right. These are hypothetical retail examples:

• A returns answer says sale items can never be refunded, including faulty goods.
• A delivery answer promises arrival before an event without a supported delivery commitment.
• A promotion answer invents a discount code or omits its exclusions.
• A product answer says a jacket is waterproof when its documents only describe water resistance.
• A stock answer says a size is available when the inventory shows it has sold out.

A fluent answer can still contain an invented fact. Read the guide to AI hallucinations for the cause. For the actual refund and repair rules, use the consumer guarantees guide.

Origin and Green Claims in Chat

Environmental and product origin claims need the same care in chat as on a product page. Terms such as ‘eco-friendly’ or ‘carbon neutral’ can mislead when the evidence does not support the impression they create.

In 2025, the Federal Court ordered Clorox to pay $8.25 million in penalties. Certain GLAD bags carried misleading ocean-plastic claims. Clorox admitted the products did not contain the claimed proportion of plastic collected from the ocean or sea.

Food origin terms have different legal meanings. Under the ACCC’s country of origin guidance, ‘Made in Australia’ concerns the last substantial transformation of the product. Imported ingredients can be present. ‘Product of Australia’ requires the main ingredients to come from Australia and almost all processing to occur here.

Most retail food needs a country of origin label, including food sold online. Shoppers see it on packaged food or accompanying information for covered unpackaged food, such as an in-store sign. The ACCC’s food labelling guidance explains the required formats. Priority foods made in Australia generally have a kangaroo logo, an ingredient-percentage bar chart and explanatory text.

Make the chat answer repeat the food’s actual label wording, including any Australian ingredient percentage. An Australian brand or local packing location cannot substitute for that label. For example, ‘Made in Australia from imported ingredients’ must not become ‘Australian ingredients’ in a reply.

Limit origin and environmental answers to approved claims in the catalogue feed and product documents. Preserve qualifications about what the claim covers. Give the agent no permission to add an origin, certification or environmental benefit from general knowledge. If the required product evidence is missing, send the question to staff.

Controls That Limit Chatbot Risk

Limit answers to current product data and policy text, with a way to trace each answer to its source. Remove superseded policy documents so the agent cannot choose an older rule over the current one.

Set topic limits that stop the agent making unsupported promises. A delivery estimate must retain its conditions, and a product recommendation must retain the documented claim. The guide to AI guardrails explains how to restrict what an agent can say.

When no source supports an answer, have the agent decline to guess and offer a person. Route refund decisions, disputes and complaints to staff who can assess the circumstances. Build a handover to staff that carries the conversation with it, so the shopper does not have to repeat the problem.

Keep conversation records that show the exact question and reply, alongside the source version used. Restrict access and set a retention policy for those records. They help you identify what a shopper was told and whether the same error affected others.

Before switching the agent on, ask realistic questions about faulty goods, missed delivery estimates and missing origin information. Check that unsupported claims produce a handover instead of a promise.

When the Chatbot Gets It Wrong

Correct the statement promptly and assess whether to honour the promise or provide another appropriate remedy. A staff member must consider what the shopper relied on and any rights that apply. Do not send the shopper back to a policy link as the entire response.

Preserve the conversation and explain the correction directly to the shopper. Fix the source content or answer rule that caused the error, then repeat the same question to check the reply. Review similar conversations to find other shoppers who received the same advice.

If the reply concerns disputed rights, financial loss or an unsupported green claim, get legal advice on the remedy and any wider corrective action. Keep the affected topic under staff control until the agent gives the supported answer consistently.