The chatbot on the website of an aesthetic treatment provider was only meant to book appointments and answer questions. When asked, it introduced two of the practice’s doctors as specialists in plastic and aesthetic surgery. They held no such specialist titles. The consumer protection organisation of North Rhine-Westphalia took the company to court.
On 12 May 2026, the Higher Regional Court of Hamm ruled in the consumer organisation’s favour. The company had argued that the AI’s incorrect answers could not be attributed to it. The court disagreed: a chatbot is not a “third party” but a tool of the company. According to reports on the ruling, it makes no difference whether the error stems from the programming or from the AI itself.
My view: the delay to parts of the EU AI Act sounds like breathing space. For chatbots that talk to guests, it buys little. Who has to answer for what they say is already governed by existing law, and the first rulings have gone against the businesses.
What the courts have said so far
The Hamm case concerns unfair competition law, specifically misleading statements under the German Act against Unfair Competition, with the same consequences a false advertising claim can trigger: a formal warning, an injunction, legal costs. The court had granted leave to appeal to the Federal Court of Justice because the question is new. The state’s official case law database now records the ruling as final.
A widely cited decision comes from Canada. In February 2024, Air Canada had to pay a customer around 650 Canadian dollars in damages, or a little over 812 dollars including interest and fees. Its chatbot had wrongly explained a bereavement fare discount to him. The airline argued that the chatbot was responsible for its own statements. The Civil Resolution Tribunal in British Columbia, which handles smaller civil disputes, rejected that: it makes no difference whether the information comes from a static page or a chatbot. That the correct rule appeared elsewhere on the website did not help Air Canada.
Is a hotel in Germany bound by a price or cancellation policy that its chatbot states incorrectly? In my research I found no ruling on that. Lawyers see several possible routes, through contract law or through pre-contractual liability. In both cases described here, though, the companies did not get away with blaming the chatbot.
For its AI Opportunity Study, sponsored by several technology providers, the consultancy h2c surveyed 113 hotel chains. Asked about letting AI make decisions on its own when dealing with guests, 56% name unclear liability for errors as a concern, the concern cited most often on this question.
What the AI Act requires after the delay
On 27 July 2026, the so-called Digital Omnibus, an amendment to the EU AI Act, came into force. What it means for a hotel comes down to three points.
AI disclosure has applied since August. Since 2 August 2026, AI systems that interact directly with people must be designed so that those people know it, at the latest at the first interaction, unless it is obvious anyway. This obligation falls first on the provider of the system, and it was not postponed. For a hotel, that means checking whether its own chatbot actually displays the notice.
The AI literacy duty has been softened. Since February 2025, businesses had to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff. Now they must take measures to promote that literacy. The law no longer names a specific level.
The high-risk rules have been postponed. For the areas listed in Annex III of the Act, they now apply from 2 December 2027 rather than August 2026. For hotel groups, this mainly concerns HR: AI that filters job applications or assesses employees’ performance and behaviour can fall under these rules.
Then there is Germany’s new Product Liability Act, passed by the Bundestag on 8 October and due to come into force on 9 December 2026. It expressly covers software, including AI. It provides for compensation for personal injury, damage to property not used exclusively for professional purposes, and the destruction or corruption of data not used for professional purposes. As I understand it, an incorrect price quote is not covered. For hotels, the law is mainly a reason to review their contracts with their own suppliers.
The case against
Reading these rulings, you might be tempted to switch the chatbot off altogether. But incorrect information is not unique to AI. An outdated cancellation page or a temp who promises a goodwill gesture on the phone can also cause trouble, and the legal position differs from case to case. A chatbot adds scale: it can repeat the same mistake across many conversations before anyone notices.
That is why I think doing without it is the wrong answer. Better to decide what the AI may talk about, where it gets its answers from and when a human takes over.
What I would advise
1. Answer only from approved sources. Property information, cancellation terms, prices from the PMS. No open web searches, no answers from the model’s memory.
2. Draw boundaries. Goodwill gestures, discounts, group rates, allergens, individual accessibility questions and anything legal go to a person. The chatbot says so openly and hands over.
3. Check the notice. Does the guest see at first contact that they are writing to an AI? That applies to the chatbot on the website as much as to any AI that replies automatically on WhatsApp or by email.
4. Spot-check conversations. Regularly read a handful of conversations, as far as data protection rules allow. Some errors can then be fixed at source.
5. Clarify who is responsible. In groups, sales and legal should decide together what the chatbot may say. In an independent hotel, one person in charge is enough. And supplier contracts should state who does what when errors occur.
What I would not do: rely on a line like “AI can make mistakes” as a get-out clause. Whether such a notice helps in court is an open question. For the court in Hamm, at any rate, it did not matter how the error came about.
One standard for every channel
My standard would be this: the front desk, the website and the chat window give the same information. A chatbot should therefore answer only what it knows from verified sources and pass everything else to a person.
That is how we at 5213 LABS would set up AI for dealing with guests: tight boundaries at first, widened step by step. This article is an assessment, not legal advice; for individual cases, please consult a lawyer.