Google loses host privilege for AI Overviews on three grounds in Munich

The written reasoning of the Munich Regional Court docket I judgment of Could 28, 2026, in case 26 O 869/26, rests on three arguments that collectively dismantle the legal responsibility protect Google has held in Germany because the early 2000s: AI Overviews converse in their very own phrases, they include claims present in no supply, and treating them like search outcomes would depart victims of machine-generated falsehoods with none authorized treatment.

The end result of the case is thought. PPC Land reported in June that the twenty sixth Civil Chamber issued a preliminary injunction prohibiting Google from repeating false claims its AI Overviews generated about two German publishing corporations, backed by fines of as much as EUR 250,000 per violation. The licensed copy of the judgment, delivered after an oral listening to on April 23, 2026, now permits a more in-depth studying of how the court docket reached that end result – and the place the identical reasoning might carry future disputes over generative search.

The stakes of the doctrinal query exceed this single case. For twenty years, German courts, following the Federal Court docket of Justice, handled search operators as oblique interferers: liable solely after notification of an apparent infringement, and by no means obligated to examine content material proactively. The Munich chamber declined to increase that framework, holding that the show of the contested “outcomes with AI” is “not a mere show of search outcomes, however relatively its personal content material attributable to it.” All the things else within the judgment flows from that classification.

The primary floor: a voice of its personal

In keeping with the judgment, the decisive distinction between a outcomes web page and an AI Overview begins with language. Search outcomes seem as hyperlinks or snippets. The overviews at subject summarized and offered the outcomes “within the web site’s personal phrases and in keeping with its personal construction,” opening with an affirmation of the question itself: “Sure, Verlagshaus24 (GeraMont Verlag) is thought for doubtful enterprise practices…” in a single model, and “Sure, there are indications of scams and doubtful practices in reference to Verlagshaus24…” within the different.

The court docket discovered that this affirmative opening “linguistically goes past the mere presentation of hyperlinks.” The texts then imposed a thematic structure present in not one of the cited sources: an introductory abstract, a compilation beneath headings the judgment renders as “Traits of the alleged rip-off” and “Typical Scams,” and shutting suggestions beneath “What You Can Do” and “What You Ought to Do.” All of this, the chamber wrote, demonstrates an impartial, content-based processing of the search outcomes by the AI provided by the respondent. As a result of Google launched the system and alone controls its algorithms, it should reply for what the system produces.

The second floor: claims that exist nowhere else

The argument the court docket described as weighing “above all” considerations fabrication. In keeping with the judgment, not one of the hyperlinks displayed alongside the overviews established any connection between the publishing corporations and the third-party companies the AI related to them, nor did any supply declare frequent identify modifications between the entities. The statements originated within the overview itself.

The sourcing element is stark. Each overviews cited, in first place, an article by the legislation agency Loschelder Leisenberg that “incorporates no reference by any means to the plaintiffs.” An affidavit from the Affiliation of German Publishers and Booksellers, submitted as Exhibit ASt 14, established that no connection exists between the publishers and the businesses named within the AI textual content. Google didn’t contest this with substantiated argument. On that file, the court docket concluded the overview constitutes “a separate assertion generated by the AI supplied by the respondent, for which the respondent, because the supplier (additionally inside the that means of Artwork. 3(3) of the AI Regulation), should settle for duty.”

The reference to Article 3(3) of the EU AI Act is a quiet however notable transfer. The court docket didn’t apply the regulation as a foundation for the declare – it expressly held the AI Act affords solely a grievance path to market surveillance authorities – but it borrowed the regulation’s supplier idea to anchor duty for system output in nationwide civil legislation.

The third floor: a niche nobody else can fill

The chamber then requested what would occur if AI-generated statements obtained the notice-and-takedown therapy developed for linked content material. Its reply provides the judgment’s most consequential passage. Affected events can’t sue the operators of the supply web sites, as a result of these third events by no means made the fabricated statements; the claims have been “first compiled and disseminated by the search engine operator” within the overview. If the operator then owed nothing past eradicating apparent violations, victims of statements which might be illegal solely after thorough examination “would then haven’t any sufficient alternative to acquire authorized safety.”

That safety hole, in keeping with the court docket, is itself a purpose why the precedents on search engines like google and autocomplete can’t be transferred to AI Overviews.

Distinguishing twenty years of BGH precedent

Google constructed its protection on two Federal Court docket of Justice choices: the 2018 search engine ruling in case VI ZR 489/16 and the 2013 autocomplete ruling in case VI ZR 269/12. The Munich judgment works by each.

The 2018 determination shielded search operators as a result of a proactive overview obligation “would battle with the aim and functioning of a search engine,” and since, with out search, the web “could be unusable for people as a result of overwhelming flood of knowledge.” The Munich chamber accepted that premise and denied its utility. An AI-powered overview, the court docket held, “is in no way important for utilizing the web,” since displaying hyperlinks already makes the flood of knowledge manageable. The characteristic as an alternative “constructions and evaluates information in keeping with a system not instantly recognizable to the consumer” and thereby channels the reply. Comfort, within the court docket’s evaluation, doesn’t earn the constitutional solicitude granted to necessity.

The court docket additionally disposed of the sensible argument behind the 2018 ruling: {that a} search operator has no relationship with the authors of listed pages and can’t examine their claims. Google faces no such impediment right here. The corporate “is actually able to verifying the content material of the AI Overview in opposition to the sources on which it’s based mostly, even with out contacting the third events.” In keeping with the judgment, Google failed to take action even after the publishers notified it – by legal professional letter on February 2, 2026, by electronic mail the identical day, and once more by Google’s personal on-line kind after the corporate replied that it couldn’t course of the emailed grievance.

The autocomplete precedent fared no higher. In 2013, the BGH discovered operators accountable in precept for search strategies however restricted duties to circumstances of notified infringement. Munich distinguished the merchandise: at subject is neither a question suggestion nor an inventory of hyperlinks however “an independently generated assertion that, based mostly on the weighting of the search outcomes and an analysis of the content material, formulates its personal reply (not a query) and affords it to the consumer.”

The chamber left open whether or not a overview obligation triggered solely by notification would possibly suffice for AI output, as a result of the publishers had undisputedly notified Google by two channels. It rejected, nevertheless, any limitation to apparent violations, provided that the system’s personal sources allow verification.

The front-page reader meets the AI reply

Throughout the oral listening to, Google argued that knowledgeable customers know AI-generated info shouldn’t be blindly trusted and might examine the connected hyperlinks. The court docket answered with a longtime doctrine of German press legislation. Informal readers – the judgment invokes the determine of the front-page reader, developed by the Federal Constitutional Court docket in 1998 – could take a self-contained assertion at face worth, and the potential of refuting a press release by additional analysis doesn’t relieve the speaker of legal responsibility for it.

Utilized to the product: the overview “is understandable in and of itself, incorporates a self-contained assertion with independently comprehensible content material, and makes no reference to different potential interpretations and even to any unreliability of the content material, in order that customers usually haven’t any purpose to overview the displayed reply to the search question.” The court docket added a pointed commentary about product logic. If overviews have been usually acknowledged as unreliable and each hyperlink required impartial verification, this “would considerably diminish the very utility of the characteristic cited by the respondent.”

Opinion constructed on invented info

The judgment’s second half classifies every contested assertion as truth or opinion, and the taxonomy explains each what Google misplaced and what it received.

The headline assertion – that the writer is thought for doubtful enterprise practices and sometimes perceived as a rip-off – certified as an expression of opinion, as a result of factual and evaluative parts have been inseparably intertwined. Underneath German constitutional legislation, opinions needn’t be well-founded, substantiated or convincing. The safety collapsed anyway, for a purpose the court docket situated in Federal Constitutional Court docket case legislation: the weighing should take into account whether or not an opinion rests on true info or on “an evaluation arbitrarily plucked out of skinny air.” Right here the opinion rested totally on the alleged connection to subscription-trap corporations, a connection the affidavit disproved. An opinion constructed on a false connecting truth loses the balancing.

The court docket added a consideration particular to machine output. The opinion “was primarily generated by AI, that means it isn’t an expression of a conviction held by the individuals making the assertion, however relatively the results of an algorithm.” Providing AI-supported search, the chamber discovered, is primarily industrial exercise and solely secondarily participation in public discourse – a framing that lowered the burden of Google’s expression curiosity within the steadiness.

Statements open to proof – the corporate associations, the billing after phone calls that by no means passed off, the failure to unlock paid content material, the unreachability – have been handled as factual assertions. As a result of they injury repute, the burden of creating fact fell on Google beneath the rules of Sections 186 and 187 of the German Felony Code. Buyer evaluations on score portals didn’t carry that burden: in keeping with the judgment, the evaluations involved dissatisfaction with service and order processing, not unwitting subscription sign-ups, and overview platforms don’t confirm what third events publish. Nor had Google proven that it, or the AI it employs, had subjected the statements to any overview in any respect.

The place the plaintiffs misplaced

Two claims failed, and the failures hint the bounds of the doctrine. The allegation of cooperation with a debt assortment company was, within the court docket’s view, not damaging sufficient to shift the burden of proof to Google; the publishers would have needed to set up its untruth and didn’t. And the asserted assertion that the plaintiffs “promote subscriptions and gather funds in an unethical method” seems in neither overview. At most readers would possibly infer it, and an inference fashioned by readers can’t be prohibited until it imposes itself as an inevitable factual declare. The price cut up displays the result: every plaintiff bears 10 p.c of court docket charges, Google 80 p.c.

Google’s ultimate argument – {that a} machine studying revision had eliminated the texts, eliminating repetition threat – additionally failed. With out a penalty-backed cease-and-desist declaration, and provided that recurrence “can’t be reliably dominated out” beneath the algorithms in use, the chance persevered. The injunction, the court docket held, is just not restricted to German territory, as a result of Article 36(1) of the Brussels I Regulation extends recognition of the choice throughout the regulation’s scope.

The reasoning arrived in a authorized setting that had been circling the query. A Frankfurt Regional Court decision of September 10, 2025 accepted in precept that an objectively incorrect AI Overview might represent illegal obstruction, however dismissed the case as a result of the abstract proved finally not false. Munich provides what Frankfurt lacked: a documented falsehood, and a full doctrinal pathway from fabricated output to enforceable injunction.

Commentary from the promoting trade learn the judgment by the economics of zero-click search. “This judgment hits the nail on the pinnacle,” wrote Daniel Rijo, a programmatic advertising and marketing skilled at Havas Media Germany, on LinkedIn, noting far-reaching implications together with for competitors legislation. “Economically, technically and legally, that is compelling: in the event you synthesise third-party content material and add some hallucinations to offer your complete personal reply, thereby eliminating incentives to click on by to the sources, you can’t argue that you’re merely a impartial host supplier. You might be accountable to your reply, even whether it is unsuitable.”

The press-through dimension he references is quantified. Analysis throughout 300,000 key phrases discovered organic clicks to top-ranking pages fall 34.5 percent when AI Overviews seem, and the Independent Publishers Alliance complaint of June 30, 2025 put that visitors loss earlier than the European Fee. The Munich reasoning connects the 2 grievances: the identical synthesis that removes the inducement to click on is what converts Google from middleman to writer. The European Fee is individually preparing a record Digital Markets Act fine that features considerations about AI Overviews self-preferencing, and the Court docket of Justice confirmed the 4.1 billion euro Android fine on July 2, 2026, so the legal responsibility discovering lands on an organization already defending its search mannequin on a number of European fronts concurrently.

For companies of any measurement, the doctrinal shift modifications what a false AI Overview means. It’s now not third-party content material awaiting discretionary overview however the operator’s personal speech, contestable beneath company character rights with per-violation penalties connected. The judgment stays preliminary aid and could be appealed. Its structure, assembled step-by-step from Federal Court docket of Justice and Federal Constitutional Court docket precedent, was plainly constructed with that enchantment in thoughts.

Timeline

• January 20, 2026: A search combining the writer’s identify with the time period “Betrugsmasche” returns an AI Overview affirming the corporate is thought for doubtful enterprise practices
• January 26, 2026: A second search returns the same overview; each cite a legislation agency article containing no reference to the plaintiffs
• February 2, 2026: The publishers’ legal professional sends a cease-and-desist letter and electronic mail; Google replies the identical day that it can’t course of the grievance and refers to its on-line kind, by which the grievance is resubmitted
• February 10, 2026: An extra search, after the warning letter, nonetheless yields a abstract textual content related in key respects
• February 24, 2026: Google requests additional info concerning the inquiry
• April 23, 2026: The Munich Regional Court docket I holds the oral listening to
• Could 28, 2026: The twenty sixth Civil Chamber points its judgment in case 26 O 869/26, granting many of the injunction and assigning Google 80 p.c of prices

Abstract

Who: The twenty sixth Civil Chamber of the Munich Regional Court docket I, ruling for 2 German publishing corporations – a home bundling 12 publishing manufacturers and its subsidiary working the GeraMond imprint – in opposition to Google as operator of google.de.

What: The written judgment classifies AI Overviews as Google’s personal statements on three grounds: they summarize ends in their very own phrases and construction, they include claims present in no supply, and shielding them would depart affected events with out authorized safety. The classification strips Google of the internet hosting exemption beneath the Digital Companies Act and the notice-and-takedown privilege for search engines like google, exposing the corporate to fines of as much as EUR 250,000 per repeated violation. Two contested statements survived, and Google bears 80 p.c of prices.

When: The judgment was issued on Could 28, 2026, following an oral listening to on April 23, 2026, over AI Overviews displayed between January 20 and February 10, 2026.

The place: Munich, Germany, case quantity 26 O 869/26, with impact extending past German territory beneath Article 36(1) of the Brussels I Regulation.

Why: The court docket discovered the overviews fabricated connections between the publishers and fraud-linked corporations, cited a supply that by no means talked about the plaintiffs, offered self-contained solutions that front-page readers haven’t any purpose to confirm, and rested a dangerous opinion on unfaithful connecting info – making the texts Google’s personal speech beneath German company character rights legislation.