Generative AI is rapidly becoming a standard part of business strategy. Chatbots increasingly serve as the first line of customer support, product advisers, or query-handling tools. From a business perspective, the appeal is obvious: a chatbot operates around the clock and handles high volumes of interactions. The problem arises when it gives a wrong answer.
From the consumer’s perspective, the output is not an abstract technological artifact. The consumer is interacting on the entrepreneur’s website or in their application. If the entrepreneur places a chatbot in the purchasing journey, the consumer will typically perceive its responses as statements of the entrepreneur, without distinguishing whether the information was produced by an employee, an automated rule, or a language model.
This view is consistent with European consumer materials. The European Consumer Centres Network has flagged risks including irrelevant answers, misunderstood queries, and insufficiently specific assistance. Dutch supervisory authorities have similarly warned that consumers may find themselves passing through one chatbot after another without knowing whether they are dealing with a human or a machine, while emphasising that the chatbot speaks on behalf of the organisation. The entrepreneur should therefore maintain control over the information the chatbot provides, and errors should be easily remedied through human contact.
The relevant question is therefore not whether generative AI can make mistakes. It can and does. What matters is who bears the consequences when an AI chatbot errs in commercial communication with a consumer. Available foreign case law points firmly toward the entrepreneur.
Moffatt v. Air Canada (2024)
In this British Columbia Civil Resolution Tribunal case, a customer was told by the airline’s chatbot that a bereavement fare could be claimed retrospectively within 90 days of travel. The customer acted on this information, only to find that Air Canada did not in fact permit retrospective claims. The correct policy appeared elsewhere on the same website.
Air Canada argued it could not be responsible for the chatbot’s output. The Tribunal rejected this, noting that the company was in effect arguing the chatbot was a separate legal entity responsible for its own actions, an argument the Tribunal characterised as extraordinary. The chatbot was part of Air Canada’s website, and it made no difference whether the incorrect information came from a static page or an interactive chatbot. A consumer cannot be required to cross-check one part of an entrepreneur’s website against another, nor to determine for themselves which of the entrepreneur’s communications to trust.
OLG Hamm, ref. 4 UKl 3/25
The Hamm Court of Appeal addressed a case involving a company offering minimally invasive aesthetic procedures. Its website chatbot told prospective patients that the company’s physician-directors held specialist qualifications in plastic and aesthetic surgery. This was false: the doctors in question held no such qualifications, and some of the titles the chatbot used did not correspond to any recognised specialty.
The company argued that chatbot responses could not be attributed to it as its own commercial conduct, citing the system’s autonomous, probabilistic operation and the impossibility of identifying why any particular output was generated.
The court rejected this. Operating from a broad concept of commercial conduct encompassing technically supported or automated behaviour, it held that what mattered was that the chatbot ran on the company’s website, responded in connection with the company’s services, and served the company’s commercial interests. As in Moffatt, the court declined to treat the chatbot as a third party for whose conduct the company bore only conditional responsibility. The company had sufficient control over the chatbot’s operation and configuration, as demonstrated by the fact that, once the problem was raised, it was able to deactivate and reconfigure the system promptly.
The court also addressed the so-called black-box problem, accepting that it may not always be possible to determine retrospectively why a generative AI system produced a particular output, but holding that this does not prevent the erroneous output from being attributed to the company. Equally, the court rejected the argument that consumers are generally aware of AI unreliability and therefore do not trust chatbot responses, finding no such general rule of experience; some consumers may in fact place particular trust in computer-generated answers precisely because they perceive them as accurate or neutral.
Implications for entrepreneurs
An entrepreneur deploying an AI chatbot in consumer-facing communication should assume, in line with international trends, that chatbot outputs may be treated as part of the entrepreneur’s commercial communication, carrying the entrepreneur’s own liability.
Under consumer protection law, particular risk arises where a chatbot provides information that is false or capable of misleading the consumer: statements about technical specifications, compatibility, certifications, return policies, discounts, or delivery timelines that could influence a purchasing decision may constitute a misleading commercial practice. The risk is not confined to express falsehoods; material omissions or ambiguous answers create equal exposure. A response that “returns are free” without specifying that this applies only to certain goods, registered customers, or unused items is a clear example.
Entrepreneurs cannot rely on a general assumption that consumers know AI systems make mistakes and will independently verify responses against the terms and conditions. The more deeply a chatbot is embedded in the purchasing journey, the less persuasive it becomes to argue that a reasonable consumer should not have relied on its output without further verification.
Conclusion
An AI chatbot can be a valuable business tool. Legally, however, it is not a neutral instrument. Once deployed in consumer communication, its outputs may be treated as part of the entrepreneur’s own commercial conduct. Available case law and European soft-law materials point consistently in the same direction: an entrepreneur cannot present a chatbot as customer support or a sales assistant while simultaneously disclaiming responsibility for its responses.