Aesthetify's chatbot invented doctors' titles. A court held it liable

A German beauty clinic's website bot called its two doctors board-certified surgeons. They are not. The clinic fixed it in days and still lost in court.

Aesthetify GmbH, a beauty-treatment clinic in Recklinghausen, Germany, put an AI chatbot on its website to answer patient questions and book appointments. On 3 April 2025 it told a user that the clinic's two managing doctors were specialists in plastic and aesthetic surgery, which they are not. On 12 May 2026 the Higher Regional Court in Hamm ruled that every word the bot says is the company's own statement, and ordered it to stop, under threat of fines of up to EUR 250,000 per breach.

The money in the case is small. The ruling is not. It is the clearest written answer yet to a question every company running a customer-facing bot will eventually ask: when the machine makes something up, whose statement is it?

What ran, and for how long

The bot answered visitors in real time and offered to book them in. It was built, configured and trained for the clinic by an outside consultancy, which the judgment identifies only as D Consulting GmbH. According to the clinic, it was trained only on the clinic's own website text and FAQs, and the court accepted that none of that material claimed the two doctors were specialists. The clinic's site correctly advertised specialist titles only for other doctors who held them.

Asked three versions of the same plain question, the bot gave three confident wrong answers. Yes, they are specialists in plastic and aesthetic surgery. They are specialists in aesthetic medicine. They are specialists in aesthetic treatments. The last two titles do not exist as specialist qualifications in Germany. Each answer ended by offering an appointment.

The judgment does not say when the bot went live. It records how fast it came down: a consumer association, Verbraucherzentrale Nordrhein-Westfalen, sent a warning letter on 10 April 2025, and the clinic switched the bot off immediately. It relaunched it with a prompt instruction to answer neutrally on any question containing the German word for specialist doctor, and a filter that strips the word from replies. On 23 April 2025 the clinic admitted the answers were wrong, in writing, and called them a regrettable technical oversight.

What did it cost?

On paper, very little. The association asked for EUR 260 to cover its warning letter and a signed promise, backed by a penalty, not to repeat the claims. The clinic refused both, because it had already fixed the bot. The court then ordered the EUR 260 with interest, put all the costs of the case on the clinic, and replaced the promise it would not sign with a court order backed by fines of up to EUR 250,000 per breach.

The fix took days. Refusing to sign for it took a year and a published judgment.

That is the operator's lesson in one line. Under German unfair-competition law, a fix does not end the risk of repetition in the court's eyes; a signed, penalty-backed undertaking normally does. The clinic had the cheap exit and declined it.

Why the defences failed

The clinic made the arguments any company would make, and the court took each one apart.

The bot acts on its own, so its words are not ours. The court said the bot is a tool the company chose to deploy and could steer, and the proof was the fix itself: if a prompt rule and a keyword filter stopped the false answers in days, the company had control all along.

We could not have foreseen this question. The court said anyone selling cosmetic procedures knows patients ask about qualifications, and that health is an area where the standard of care is high. A careful business would have tested exactly that question before launch.

Our vendor built it on accurate content. Irrelevant. German law attributes a contractor's conduct to the business that hired it.

Users know chatbots get things wrong. The court found no evidence for that, and noted that the clinic would hardly have installed a bot its customers did not trust.

The court left one door open. It did not decide what happens when a user deliberately leads a bot into a false answer. These questions were neither leading nor suggestive.

Does this apply to a 200-person company outside Germany?

The legal route differs by country, but the answer so far does not. In February 2024 a Canadian tribunal rejected Air Canada's argument that its website chatbot was responsible for its own actions and ordered the airline to pay CAD 812.02 after the bot misstated a refund rule. If your bot answers customers, plan on its words being yours.

What this does not mean

It does not mean every chatbot error ends in court. This case reached a judge because a consumer association, which in Germany can sue on its own initiative, found a false claim about professional qualifications in a health business, and because the clinic would not sign a EUR 260 settlement. The case was about the statements themselves; the only money ordered was the association's EUR 260 fee, interest and costs.

It also does not mean you must switch your bot off. The court's own reasoning tells you what reasonable care looks like: think about which questions your customers are certain to ask, decide which wrong answers would be unlawful or costly, and test those before launch.

What to do this week

List the ten questions where a wrong answer from your bot would be a false statement you could be held to: staff qualifications and certifications, prices, refund and cancellation terms, delivery promises, warranty cover, safety claims. Ask your bot each one, several ways, and keep the transcript. Where it guesses, block the topic or hand it to a person, as Aesthetify did after the fact.

And if a warning letter arrives, fix the bot and take the cheap exit. Aesthetify did the first part straight away. The second part is what put its name in a court press release.