Incorrect chatbot answers: liability as for your own advertising

Invented specialist medical titles in the chat

On the website of a provider of aesthetic treatments, customers could communicate with an AI chatbot, book appointments and ask questions in real time. In response to specific questions, the chatbot stated, among other things, that the two doctors behind the business were „specialists in plastic and aesthetic surgery“, „specialists in aesthetic medicine“ and „specialists in aesthetic treatments“. The two managing directors do not hold these specialist titles. It was undisputed in the proceedings that the statements were false.

The consumer advice centre of North Rhine-Westphalia issued a warning letter to the company and demanded a cease-and-desist declaration with a penalty clause, that is, a promise not to repeat the statements, backed by a contractual penalty. The company switched the chatbot off but did not sign, and the consumer advice centre went to court. The Higher Regional Court of Hamm upheld the claim in its judgment of 12 May 2026 (case number 4 UKl 3/25) and ordered the company to stop using certain specialist medical titles. Since July 2025, the court has had central jurisdiction over claims under the Injunctions Act (Unterlassungsklagengesetz) for the whole of North Rhine-Westphalia.

Switching it off did not end the dispute

The course of events alone contains a lesson, because the company had deactivated the chatbot after the warning letter and was still found liable. In competition law, it is not enough simply to stop the conduct complained of. Anyone who has once advertised in a misleading way must eliminate the risk of repetition, and as a rule that is only possible with a cease-and-desist declaration backed by a penalty. Anyone who refuses gets the prohibition by judgment, even if the chatbot has long been switched off. This applies to software that can be deactivated with a click just as it does to an advertisement that someone takes offline.

Chatbot answers are the company's own statements

The core of the judgment is attribution. The company had argued that the false answers could not be attributed to it as its own commercial practice. In competition law, a commercial practice is any statement by which a company seeks to win or retain customers, advertising being the classic example. The panel did not follow the company. The chatbot's answers are unlawful commercial practices of the company itself: untrue statements about the trader's competence, status and authorisation, which the Act against Unfair Competition expressly prohibits.

The same standard therefore applies to the chatbot as to every advertisement, every flyer and every page of a company's own website: the content must be correct. The fact that software formulates the sentence in real time does not change the responsibility.

Correct data sets offer no protection

The panel is clear on the question of whether a company can exonerate itself with clean technology. Even if it had the chatbot programmed exclusively with correct data sets, it bears responsibility for the false statements. The risk that a language model will form incorrect sentences out of correct information therefore lies with the operator who deploys the chatbot.

The detour via the rules on third-party content is also blocked. Anyone who allows third-party content on their platform, for example reviews or advertisements by other users, is only liable if they breach reasonable duties to check. Lawyers speak of duties of care. The panel, however, does not regard the chatbot as a third party within the meaning of the law, so such duties to check do not come into play at all. Its answers form part of the company's own presence, like the texts on the website.

Liability for chatbot answers in practice

The decision affects every company that lets a chatbot answer questions on its website, from the online shop and the medical practice to the law firm. Everything the chatbot says about the company, its services and its qualifications counts as a statement by the company. Mistakes have the usual consequences under competition law: a warning letter, a cease-and-desist declaration and, in case of dispute, litigation. Competitors and associations such as the consumer advice centre can assert these claims.

Statements about titles, qualifications and authorisations are particularly sensitive, because the law expressly prohibits misleading statements about precisely these matters. Suppose a law firm without a certified specialist lets a chatbot answer enquiries from prospective clients, and on request it calls the owner a „certified specialist in employment law“. Following the line taken in the Hamm judgment, that would be misleading advertising by the firm and not a mistake it could pass off on the software.

Test the chatbot and limit its topics

What does this mean in practice? The panel did not supply a checklist, but the direction follows from the reasoning. If correct data sets alone do not exonerate you, the operator needs to know what their chatbot actually answers. It makes sense to test the system before launch and regularly thereafter with the questions customers ask: who is behind the company, which titles and authorisations exist, which memberships. If the chatbot invents something, the error should be fixed before a customer reads it.

It is equally sensible to limit the chatbot to what it reliably masters. Booking appointments is something quite different from giving information about specialist medical titles. One solution is to take questions about qualifications away from the chatbot and have it refer to a vetted page or a member of staff. That reduces the risk precisely where it arose in the Hamm case. And if a warning letter arrives anyway, the matter does not end with someone pulling the plug.

Appeal to the Federal Court of Justice allowed

The last word may not yet have been spoken. Because the attribution of false statements made by an AI chatbot raises new legal questions, the panel allowed an appeal on points of law to the Federal Court of Justice (BGH). Whether the company will take that route remains open. The detailed reasoning is set out in the written grounds of the judgment, which the state publishes in its case law database. For operators, that changes little: the question of what your own chatbot tells customers arises regardless of how Karlsruhe ultimately decides.