German court makes Google liable for AI Overview errors

A Munich court has quietly redrawn the map of AI liability. In a ruling reported by Search Engine Journal, a German court found that Google's AI Overviews constitute Google's own speech, not a neutral aggregation of third-party content. The distinction sounds technical. Its consequences are not.

Under the old framework, Google operated as a conduit: it surfaced links, ranked pages, and disclaimed responsibility for what those pages said. AI Overviews break that model entirely. The system does not point to a source; it synthesises one. When the synthesis is wrong, there is no third-party publisher to absorb the blame. The Munich court concluded, logically enough, that if you write the sentence, you own the sentence.

What changed in Munich, and why it travelled fast

The ruling applies German law, but it signals a broader trajectory. The EU's AI Act and Digital Services Act are both premised on the idea that generative outputs carry a different class of risk than retrieval outputs. A court finding that aligns AI-generated text with editorial speech rather than search infrastructure will be cited in future proceedings across the bloc. Google is not the only company exposed: any platform that presents an LLM-generated summary as an answer, rather than a list of sources, faces the same structural argument.

For Google specifically, the operational implication is serious. AI Overviews now run at scale, covering queries across health, finance, legal topics, and public figures. Each wrong answer is, under this logic, a potential publication tort. The company will either have to accept that liability, narrow the categories of query where Overviews appear, or engineer citation practices aggressive enough to shift responsibility back toward source publishers.

That last option has direct consequences for B2B brands.

The citation arms race gets a legal dimension

Until now, the incentive to appear in AI Overviews was about visibility: being cited meant being seen when a prospective buyer asked a general question. The Munich ruling adds a second dynamic. If Google now has legal exposure for inaccurate AI-generated claims, it has a structural incentive to favour sources that are authoritative, attributable, and demonstrably correct. Vague, unsourced, or inconsistently maintained content becomes a liability for the platform. The model will be nudged, by legal logic if not by engineering decree, toward sources that reduce Google's exposure.

For a multilateral institution like UNDRR or a standard-setting body like ISO, this is an opening. Their published materials are citable, stable, and carry institutional authority. They are precisely the kind of source a legally exposed AI system should prefer. The challenge is making that content legible to retrieval systems: structured, consistently updated, and written to answer the class of questions AI Overviews are likely to synthesise. Institutions that have treated their web presence as secondary to their PDF archives are carrying a visibility deficit that just became more expensive to ignore.

Financial services firms face the inverse problem. Regulatory constraints often prevent them from making strong, unhedged claims in public content. Yet AI Overviews, by their nature, flatten nuance: they present a synthesised answer rather than a qualified one. A firm whose cautious, compliance-shaped content is misrepresented in an Overview now has a potential remedy under the Munich precedent. That is cold comfort if the damage to a product's reputation has already propagated through however many model-generated answers.

The practical read

Content accuracy is no longer just an SEO hygiene question. It is an input into a legally sensitised retrieval system. Brands that publish authoritative, well-sourced, up-to-date material reduce the gap between what an AI Overview says about them and what is actually true. They also, incidentally, make themselves the kind of source that a liability-conscious platform wants to cite.

Google will appeal or adapt. But the legal principle, that a generative synthesis is speech, is harder to reverse than a product decision. The brands that treat their content as a source of record now, rather than a channel for messaging, are building the kind of presence that holds up in both editorial and legal scrutiny.