Can courts hold Google liable for AI Overview false claims?

The Munich Regional Court I issued a temporary injunction (case no. 26 O 869/26) holding Google directly liable for false claims made by its AI Overviews about two Munich-based publishers, who had been wrongly tied to "scams, subscription traps, and shady business practices." The court's reasoning rests on how it classifies the output: an AI Overview is "its own content, not just a list of search results," so Google is a "direct infringer" rather than merely a conduit for third-party pages. The overview "rewrites and judges results 'in its own words and according to its own structure,'" and in the case at hand it opened with "confident claims like 'Yes, [company] is known for dubious business practices'" before building out red flags and user tips — none of which, the court found, appeared in any linked source. The court called these fabricated connections "the defendant's own statements."

The court's key move is distinguishing AI Overviews from the search-engine case law that previously shielded Google. Germany's Federal Court of Justice (BGH) had held that regular search engines and autocomplete carry only indirect-infringer liability, because they merely make third-party content findable and a duty to proactively check results "would threaten how search engines work." The Munich court found that reasoning doesn't transfer: a search engine "just points to outside websites," while an AI Overview generates "independent, new, and substantive statements" by evaluating and combining multiple sources, and "only Google can check those statements... by comparing the underlying third-party websites with its own statements based on them." The court also rejected Google's defense that users "could check the linked sources themselves," finding the overview "understandable on its own" with "no reference to other possible interpretations," drawing a parallel to press law (publishers are liable for self-contained teasers even if nobody reads the full article), and citing studies that users "almost never click on sources in AI overviews." It further held AI output isn't protected as free speech, since an AI's "opinion" is "not the expression of an acquired conviction... but the result of an algorithm," and that Digital Services Act host-provider protections don't apply because Google isn't hosting third-party content — it's generating its own.

This is a different liability theory from the two prior chatbot rulings in the library. Can companies escape chatbot liability through careful training? and Can a company escape chatbot liability by calling it separate? both treat the chatbot as a tool of the company — not a separate "third party" — so the company answers for its output regardless of training quality, under unfair-competition and negligence doctrine respectively. The Munich court reaches the same practical result (Google can't disclaim responsibility) through a narrower route: it carves an exception out of the BGH's existing search-engine liability shield by distinguishing what kind of content each product generates, rather than applying a general "tool of the operator" rule. The court's observation that users rarely click through to sources also supplies a plausible mechanism for Does AI search summaries divert traffic away from Wikipedia?: a self-contained answer that needs no verification is exactly the design the Wikipedia study's traffic drop would predict.

The ruling is a temporary injunction under German law, "not yet final," with Google able to appeal, and it says nothing about US, UK, or other jurisdictions' frameworks despite the article's claim of possible "international reach." The excerpt's broader framing — that Gemini 3 AI Overviews answered correctly "91 percent of the time" per a separate Oumi analysis for the New York Times, with 56 percent of correct answers "couldn't be backed up by the sources Google linked" — comes from a different study, not this case, and the excerpt doesn't connect the two beyond juxtaposition. What the ruling does establish, for this jurisdiction and fact pattern, is that generating new synthesized claims rather than indexing or excerpting forfeits the liability protections search engines have historically enjoyed — a distinction that, if it holds on appeal and generalizes, would bear on any system that paraphrases web content into standalone answers, including the conversational assistants the piece names (ChatGPT, Claude, Perplexity) but does not analyze.