The Chatbot With a Medical Credential Printer

Aesthetify GmbH offered minimal invasive beauty treatments through a website that included an AI chatbot. The chatbot answered visitor questions in real time and also supported appointment booking. That made it more than a novelty widget. It was part of the sales and patient communication funnel.

On April 3, 2025, the chatbot was asked about the specialist qualifications of the company's two physician-directors. According to the Higher Regional Court of Hamm's judgment, the bot answered that they were specialists in plastic and aesthetic surgery. It then went further, referring to specialist titles in aesthetic medicine and aesthetic treatments.

The problem was direct and material: the doctors did not hold the claimed specialist qualification, and two of the titles the bot used do not exist as German specialist medical titles. In healthcare marketing, professional qualifications are not decorative. Patients use them to decide who they trust with medical procedures. Inventing them is not the kind of customer-service flourish anyone wants from a probability machine with a chat bubble.

A consumer association sent a warning letter and asked for a cease-and-desist undertaking. Aesthetify disabled and modified the chatbot, including prompts and keyword filters intended to stop future "Facharzt" claims. The company did not sign the requested undertaking and did not pay the requested warning-letter costs. The case went to the Higher Regional Court of Hamm.

The Defense

Aesthetify's defense was the one every company with a chatbot will be tempted to make sooner or later: the bot did it, not us.

The company argued that the chatbot worked statistically and autonomously within its assigned operating range, rather than being controlled line by line by a human. It also argued that the false output was not traceable to a specific cause. Maybe it was the user's phrasing. Maybe training data. Maybe missing safeguards. Maybe some combination. In any event, the company said, the statements should not be treated as its own commercial conduct.

It also argued that the bot had been trained on correct materials. The court's summary says the defendant claimed the chatbot used website text and FAQ materials in a vector database, and that those materials did not contain the false statements about the doctors' qualifications. This is a familiar defense in AI cases: the source data was clean, therefore the output problem should not be attributed to the operator.

That argument is understandable from an engineering perspective and weak from a consumer-law perspective. If a company deploys a system to speak to customers on its behalf, customers experience the output as company communication. They do not see the vector database, the retrieval pipeline, the temperature setting, the prompt, or the vendor invoice. They see a website owned by the company answering questions about the company's services.

The Court's Answer

The Hamm court rejected the attempt to treat the chatbot as a third party. It held that the bot's statements were commercial acts attributable to the company because the chatbot operated on the company's website, promoted the company's services, and was deployed within a customer and patient communication process that benefited the company.

The court drew on German unfair competition law, especially the prohibition on misleading commercial acts concerning qualifications, status, approval, or awards. False specialist-title claims fit neatly into that category. The fact that a software system generated the text did not make the statement legally weightless.

The judgment also rejected the idea that black-box behavior or lack of direct human control eliminated responsibility. The company selected, deployed, and set the operating frame for the chatbot. The court noted that the company had enough influence to disable and rework the bot after the warning letter. That practical control mattered. If the operator can modify the system after it causes a legal problem, the operator cannot plausibly pretend the system is an untouchable independent actor before the problem.

The court ordered Aesthetify to stop using the contested specialist-title claims in consumer-facing commercial conduct, threatened administrative fines up to 250,000 EUR per violation or substitute detention for violations, ordered payment of 260 EUR plus interest for warning-letter expenses, and placed litigation costs on the defendant. The court also allowed revision to Germany's Federal Court of Justice, which means the legal question may still get national-level treatment.

Why the Stakes Are Higher in Healthcare

Chatbot hallucinations are irritating when they invent a refund policy. They are worse when they invent credentials in a medical setting. A professional title is a shortcut consumers use to evaluate expertise. In a cosmetic procedure context, where patients are balancing cost, trust, risk, and medical judgment, a false specialist claim can materially affect a decision.

That is why this case fits the Graveyard even though the monetary number in the judgment is small. The harm is not a giant damages award. The harm is a court finding that a customer-facing AI system made legally misleading statements about medical qualifications, and that the company deploying it was responsible for those statements.

The ruling also has a useful lesson for anyone shipping AI chat onto a commercial website: "we gave it correct data" is not the same as "it will only say correct things." Retrieval systems can combine, overgeneralize, infer, and hallucinate. The output is what the customer sees, and the output is what regulators, courts, competitors, and consumer groups will preserve in screenshots.

The Boring Compliance Work Was the Product

A chatbot that discusses regulated claims needs guardrails built around the claims, not vibes built around the homepage copy. In this case, the sensitive topic was professional qualifications. In another business it might be medical eligibility, lending terms, cancellation rights, warranty coverage, immigration advice, insurance benefits, or safety instructions.

The control work is boring: restrict topics, refuse sensitive claims unless they match a verified source exactly, route ambiguous questions to humans, preserve logs, test adversarial and ordinary prompts, and monitor outputs after launch. Boring is the point. Compliance is usually boring right up until the court order arrives, at which point everyone suddenly discovers a deep personal interest in governance.

The Hamm decision does not say companies can never use AI chatbots. It says companies cannot use a chatbot as a liability airlock. If the bot speaks for the business, the business owns what it says.