German courts hold Google liable for AI Overview errors
German courts just made AI-generated errors about your organization potentially litigable, which will push Google toward narrower, more cautious answers.
Key takeaways
- Munich and Aachen regional courts independently ruled AI Overviews are Google's own statements, not neutral summaries.
- Google settled the Munich case rather than risk an appellate ruling cementing the precedent.
- The ruling undercuts the platform-as-conduit defense that has shielded tech companies from publisher-level liability.
- Brands with thin or ambiguous public records face higher risk of being conflated with similarly-named entities in AI Overviews.
- Expect Google to narrow AI Overview claims about named individuals and organizations to limit legal exposure.
Google settled the Munich case rather than let an appeals court bless the ruling against it. That was a tactical retreat, not a resolution. Per The Decoder, the Aachen Regional Court has now applied the same reasoning in a separate dispute, this time over an AI Overview that attributed someone else's criminal offenses to a named doctor. Two regional courts, independently, have reached the same conclusion: when Google's AI summarizes the web and gets it wrong, the words are Google's, not the web's.
That distinction sounds technical. It is the whole case.
Why the source doesn't shield the summary
Google's defense in both disputes rested on a familiar logic: AI Overviews synthesize third-party content, so liability should trace back to the underlying pages, not the summarizer. German courts have rejected that logic twice now. The doctor in the Aachen case was libeled not by the original source material, whatever it said, but by Google's generated sentence conflating him with a different person entirely. The court treated that sentence as original Google speech, subject to the same defamation standards as a newspaper article or a press release. No AI disclaimer, no "results may vary" footnote, changed that analysis.
This matters because it closes off the argument tech platforms have used for two decades: we are a conduit, not a publisher. Section 230 in the US and intermediary-liability frameworks across the EU were built on that conduit theory. Generative search breaks the theory because the platform no longer links to an answer; it writes one. Two German courts have now said, in effect, that writing an answer makes you an author, and authors are liable for what they write, hallucinated or not.
The exposure this creates for named entities
For a bank, a multilateral agency, or an industrial group, the immediate risk is not Google's balance sheet. It is the brand's own name sitting inside an AI Overview that nobody at the brand wrote, reviewed, or approved. A development bank wrongly described as under sanctions investigation, a chemicals manufacturer wrongly linked to a safety recall that belonged to a competitor, a pension fund wrongly cited in a solvency dispute that never happened: these are not hypotheticals, they are the generic failure mode of summarization at scale, and the Aachen case shows exactly how it happens. The model conflates two similarly-named or similarly-situated entities and presents the error with the same flat confidence as an accurate answer.
Until now, the practical remedy for a wronged brand was limited: flag it to Google, wait, hope the snippet updates. German case law gives wronged parties something sharper: a cause of action that treats the Overview as a statement Google made, which means standard defamation and unfair-competition remedies apply, including the kind of injunctive relief that forced Google's hand in Munich. That changes the calculus for comms and legal teams simultaneously. A factual error in an AI Overview is no longer just a reputational nuisance to monitor; it is potentially litigable content, and European courts are building the precedent to prove it.