German Court Ruling Makes Google Liable for AI Overviews

The Regional Court of Munich I (Landgericht München I) issued a temporary injunction on 28 May 2026 that quietly rewrites the legal status of generative search. Case number 26 O 869/26, decided in summary proceedings by the chamber for press law and competition matters, classifies Google not as a neutral indexer of third-party pages but as the author of its AI Overviews, with direct liability for what those summaries assert [1]. The court ordered Google to stop publishing a set of false claims about two Munich-based publishing companies and to cover roughly 80% of the plaintiffs' legal costs [3]. The reasoning, not the cost-allocation figure, is what the rest of the AI search industry will spend the next year trying to digest.

The plaintiffs, both Munich-based publishers represented by Lausen Rechtsanwälte, had complained since early 2026 that AI Overviews opened with sentences such as "Yes, [company] is known for dubious business practices" and appended structured "red flags" and user tips that did not exist in any of the linked sources [1]. The AI conflated the plaintiffs with genuinely problematic firms, asserted unsubstantiated phone calls and subscription traps, and presented the resulting text as a coherent narrative summary. Google had been given a chance to remove the claims; the court found its response inadequate, and the risk of the algorithm simply regenerating the same false statements on the next query was treated as a continuing infringement [1]. The redacted judgment is publicly hosted by The Decoder [1].

Why AI Overviews are no longer "search results"

The doctrinal move in 26 O 869/26 is short and consequential. German press and competition law has, for roughly two decades, given search operators a qualified shield. The Bundesgerichtshof (Federal Court of Justice) treated them as indirect infringers, or mittelbare Störer, on the theory that they merely point to statements made elsewhere. A search result that returns a defamatory third-party page is, in that line of cases, the third party's speech, not the search engine's. The Munich chamber held that this limited liability does not survive the move from indexed links to generated prose.

AI Overviews, the court found, are "independent, new, and substantive statements" produced by an algorithm that Google alone designs, trains, and operates [2]. Because only Google can correct the model that emits the false claim, the standard notice-and-takedown logic of the EU Digital Services Act (DSA) does not apply, and the host-provider safe harbor Google might otherwise lean on is unavailable [3]. The court put it bluntly: "Google built the AI, Google offered it to users, so Google owns what it produces, because it alone has influence over the AI's offering and the algorithms with which the AI operates" [1].

The chamber also rejected the free-speech framing that has been floated in other jurisdictions. An AI's "opinion," the court wrote, is "not the expression of an acquired conviction of the persons expressing it, but the result of an algorithm" [1]. That reasoning cuts off the "pure speech" defense a chatbot maker raised in a 2025 U.S. case, which had argued that model output should be protected as the operator's own expression [4]. In Munich, algorithmic output is the operator's expression, and where it is factually wrong about a private party, the operator's commercial-speech interest gives way to the plaintiff's reputational rights [2].

Two practical points should not get lost in the doctrinal discussion. First, the ruling is a preliminary injunction issued in summary proceedings (einstweiliges Verfügungsverfahren), not a final judgment on the merits, and Google has already announced an appeal [7]. Second, the Munich chamber did not invent its conclusion in a vacuum. It explicitly built on a September 2025 ruling by the Regional Court of Frankfurt am Main, case 2-06 O 271/25, which had already accepted in principle that AI Overviews are not categorically exempt from liability, even while refusing the specific cease-and-desist request before it at that time [3]. 26 O 869/26 is the high-water mark of a German line of cases that has been developing for the better part of a year.

The accuracy backdrop the court did not have to litigate

The injunction does not turn on statistics about hallucination rates. The false statements in this case were demonstrably absent from the sources the AI cited, and that is enough. But the underlying accuracy problem is the reason the ruling will not stay confined to Munich.

A New York Times and Oumi analysis published in April 2026 found that AI Overviews running on Gemini 3 are inaccurate about 9% of the time, and that 56% of the answers that were correct still cite source links that do not actually support the claim [2]. Read those numbers together and you have a system in which the model asserts confidently, the user sees a confident answer, and the cited evidence is, in a majority of cases, the wrong evidence. A Pew Research study of March 2025 browsing behaviour, drawn from roughly 69,000 Google searches by 900 U.S. adults, found that users who encountered an AI summary clicked a traditional search result only 8% of the time, compared with 15% when no AI summary was present, and clicked a link inside the AI summary itself in just 1% of visits [8]. AI Overviews now reach more than two billion users a month [7]. If even a small fraction of those answers are defamatory, the surface area for direct liability under the Munich framework is enormous.

The Munich chamber's most damaging sentence for the defence is one that requires almost no factual record at all: "the AI overview in this case contains statements that do not appear in the search results at all" [2]. That is the proposition that turns AI Overviews from a window onto the web into Google's own statement of facts. Once that proposition is accepted, the rest of the case is just defamation.

The appeal, the regulators, and the spillover

Google's first public response, given to The Decoder, Ars Technica, and others on 10 and 11 June 2026, was restrained. "We invest deeply in the quality of AI Overviews to ensure that the overwhelming majority of responses provide accurate information," a spokesperson said, adding that the decision "is not yet final" [1][2][4]. Three days later, the company hardened its line. "This case focuses on specific and narrow errors, not the foundational way AI Overviews displays web content. We disagree with the ruling and plan to appeal" [7]. The appeal will be heard by the Higher Regional Court of Munich (Oberlandesgericht München). The narrow-versus-foundational framing is the legal argument to watch: if the appellate court treats the injunction as a fact-specific remedy, Google's global product can continue largely unchanged. If it affirms the chamber's reasoning as a general rule, the doctrinal change sticks.

The Munich ruling lands in a regulatory environment that is already closing in on AI search. In May 2026, the United Kingdom Competition and Markets Authority ordered Google to give publishers an opt-out from having their content used to power AI search features, AI Overviews included [6]. Independent publishers have a live antitrust complaint at the European Commission arguing that AI Overviews divert traffic away from the very sources they are trained on [6]. The EU AI Act's transparency obligations for user-facing AI systems take effect on 2 August 2026, although those rules govern disclosure and risk classification, not the substantive accuracy of generated text. Google is also separately appealing a November 2025 German antitrust ruling that requires it to pay roughly €572 million in the price-comparison sector [7]. 26 O 869/26 is the most pointed of these pressures, because it is the first to put a number on what an AI search provider owes the people its product misdescribes.

The doctrinal spillover is what the industry's general counsel will lose sleep over. The same Munich logic, that a generative summary is the operator's own statement because only the operator can change the model that emits it, applies in principle to Microsoft Copilot in Bing, OpenAI's ChatGPT search, Perplexity, and You.com. Each of those products rewrites, combines, and sometimes exceeds its cited sources. None of them currently face a Munich-style injunction, but the German chambers have now drawn the line, and the question for each operator is whether its retrieval and grounding pipeline is durable enough to survive the same argument that worked against Google.

What changes tomorrow

The first visible change is in tone. Forrester principal analyst Nikhil Lai, the only industry analyst on the record across the coverage, framed it directly: "AI Overviews can no longer just be helpful summaries. Now, they must be legally defendable outputs" [6]. His prediction was that operators will retreat from assertive, first-person claims and lean on hedging language of the "according to" and "some sources suggest" variety [6]. That is consistent with what the Munich chamber implicitly invited: where a claim is not directly supported by a cited source, attribute it; where it is, label the attribution clearly.

The second change is operational. Expect AI search providers to expand conservative parameter sets for German traffic, particularly for queries about named individuals, private companies, health, and financial products, and to build faster human-in-the-loop overrides for defamation complaints [3]. Expect operators to also surface more dedicated to verifiability, such as inline source quotes, persistent per-claim attribution, and a richer distinction between "the model asserts X" and "a cited source asserts X." Lai's broader point is the one to remember: "This is not a Google-specific problem. I think this will lead to the value of defensible AI, where information's verifiability and traceability become more valuable than its polish" [6].

The third change is legal. Other German press-law chambers now have a reasoned template to follow, and 26 O 869/26 will be cited in cease-and-desist letters from Berlin to Hamburg long before the appeal is decided. Operators should not treat the Munich injunction as the last word, because it is a preliminary injunction, not a final judgment, and the appellate court can narrow it. They should also not treat it as an isolated German curiosity. The Munich framework does not require a final judgment to be useful to a future plaintiff. It only requires the proposition that a generative summary is the operator's own statement of fact, and that proposition is now on the record.