A German court held Google directly liable for its AI Overviews: the Landgericht München I ruled the AI summary is Google's own speech, not a search result (26 O 869/26, since settled on appeal)

The problem

Generative search summaries state things as fact that appear nowhere in the pages they claim to summarise. In Germany that failure mode reached a courtroom when Google’s AI Overviews feature, “Übersicht mit KI”, generated a summary that linked two Munich publishers to a “Betrugsmasche” (fraud scheme) when their company name was searched alongside that word [source]. The AI Overview falsely asserted the publishers engaged in “fraudulent schemes” and “disreputable business practices”, with specifics including “subscription traps”, unauthorized invoicing for services like “company directory listings”, continued payment demands after settlement, and frequent name changes to evade identification ([source]). The Munich court itself noted the AI made claims “that are not even made in the search results” [source].

What was built

The system under adjudication is Google Search’s AI Overviews, branded in Germany as “Übersicht mit KI”, which rewrites and summarises underlying results into a generated answer displayed above the links [source]. Two plaintiffs brought the action: Verlagshaus24, a publishing house operating 12 branded imprints, and its subsidiary GeraMond, which publishes books and magazines on technology and history [source]. They sought a preliminary injunction against Google before the Landgericht München I in a case docketed 26 O 869/26 ([source]).

The outcome

On 28 May 2026 the Landgericht München I held Google directly liable, ruling that an AI overview is the operator’s own attributable content, not a mere display of search results. The decision’s published headnote, which the database marks as an editorial summary (“redaktioneller Leitsatz”) rather than the court’s own words, puts it as “Bei KI-generierten Übersichten einer Suchmaschine handelt es sich nicht um bloße Suchergebnisse, sondern um eigene, der Suchmaschinenbetreiberin zurechenbare Inhalte” [source]. The English-language analysis published alongside the full decision renders the same holding as a finding that “the disputed display of the ‘results with AI’ is not a mere display of search results, but rather content attributable to it” ([source]), and the trade press summarised it as the court declaring the “‘AI overview’ is its own content, not just a list of search results” [source]. On that basis the court issued a preliminary injunction barring Google from repeating the specific false claims, enforceable by an administrative fine (Ordnungsgeld) of up to €250,000 per violation, alternatively detention (Ordnungshaft) against management of “bis zu sechs Monaten” per instance and “insgesamt höchstens 2 Jahre” in total [source]. Independent coverage confirms the €250,000-per-violation ceiling and the custodial alternative ([source]), and the German legal press reported the same penalty, that “Das Gericht setzte für jeden Fall der Zuwiderhandlung ein Ordnungsgeld von bis zu 250.000 Euro fest” [source]. The court allocated 80 percent of the costs to Google and 10 percent to each of the two plaintiffs, reflecting their partial success [source], a split corroborated by the trade press ([source]). The ruling was reported as “a first” in establishing this liability standard for AI-generated content [source].

What this is not

This was a preliminary-injunction proceeding (einstweilige Verfügung), not a final judgment on the merits, and it no longer has effect. The court’s published page now records that Google appealed (Oberlandesgericht München, 18 U 1744/26 Pre e), in its “Weiterführende Hinweise”: “Die Parteien haben nach Einlegung der Berufung durch die Verfügungsbeklagtenseite (Az. 18 U 1744/26 Pre e) den Rechtsstreit durch Vergleich beendet. Das Urteil zum Az. 26 O 869/26 ist damit gegenstandslos”: the parties settled after the appeal was filed, and the judgment is therefore without effect [source]. What stands is the record of what a German court held on 28 May 2026, not a ruling in force. The weakest load-bearing source here is the Transparency Coalition page: it is an advocacy organisation rather than neutral press, so it is used only to corroborate the parties and the holding, never as sole authority, with the operative text taken from the official court document and the independent trade press ([source]).

Related case files

• Google’s AI Overviews telling users to put glue on pizza and eat rocks, the same feature’s earlier, non-adjudicated factual failures
• Getty Images v Stability AI, a court drawing the limits of liability for a generative-AI system
• Spain’s AEPD ordering Worldcoin to halt, a European regulator constraining an AI deployment