Publishing, not search: German AI Overviews ruling

A German court has told Google something the AI industry has worked hard to avoid hearing: generated answers are not the same thing as search results.

Traditional web search points users towards third-party speech. AI Overviews combine, rewrite and present information as a new answer in Google’s own voice. In this preliminary ruling, the court treated the output less as a list of links and more as a substantive statement made by Google itself.

The ruling is preliminary, and Google will appeal, but the reasoning matters far beyond search. The chatbot on your website, the assistant in your CRM, and the agent drafting customer replies all produce new statements. If one of those statements is false or damaging, the question becomes less about whether the model hallucinated and more about who is legally responsible for what it said.

Google argued that users should verify AI-generated answers using disclaimers and linked sources. The court was unconvinced. A product that sells itself as providing useful answers cannot expect users to check every claim independently. If everything must be verified, much of the product’s value disappears.

For SMEs, this matters in two places. First, procurement: contracts should allocate liability for false, defamatory or commercially damaging outputs, not just IP and data protection. Second, deployment: AI-generated claims about candidates, suppliers, customers or competitors should be reviewed before they are published or acted upon.

Courts are beginning to separate AI features from the systems they sit inside. The protections available to a search engine or software platform may not automatically extend to the generated statements produced by its AI layer.

The industry has spent years insisting these systems produce original, fluent answers. Courts may simply reply: then the words are yours.

Thanks to @Douwe for the help with my personal brand.