Munich Court: Google Liable for False Answers in AI Overviews

The Munich Regional Court has ruled that Google is directly liable for false statements in its AI Overviews, in a May 28, 2026 decision that reclassifies AI-generated search summaries as the search engine’s own speech rather than third-party content. Two Munich-based publishers brought the case after AI Overviews falsely tied their companies to fraud, subscription traps, and questionable business practices, claims the court found Google alone had produced. The 26th civil chamber at LG Munich I classified Google as a ‘direct disturber’ under German press law, ordered the company to remove the specific false claims, and assigned 80 percent of the legal costs to Google. The chamber said the reasoning may have international reach, and Google, which said on June 12, 2026 that it will appeal, called the case focused on ‘specific and narrow errors.’

Munich Just Classified AI Overviews as Google’s Own Words

The 26th civil chamber at the Munich Regional Court, the division that handles press and defamation cases in the Bavarian capital, issued its ruling on May 28, 2026. Two Munich-based publishing companies brought the lawsuit after AI Overviews mixed their businesses with those of other, genuinely problematic firms. Running the case in a summary proceeding, the court issued a temporary injunction barring Google from spreading the specific false claims in further AI Overviews, with penalties if the company fails to comply, as documented in the LG Munich I order and its reasoning.

At the heart of the ruling is a single legal reclassification: Google is a direct disturber, not the indirect service provider that decades of German search-engine case law have allowed it to be. A conventional search engine only links to third-party content, but AI Overviews summarize, evaluate, and present information in a structured, original way, the court said, so the result is Google’s own statement rather than a pointer to someone else’s. Under that logic, the Federal Court of Justice’s existing shield for search engines, which protects operators from liability for the content they merely list, does not extend to AI-generated summaries. The note ‘created with AI’ attached to Overviews does nothing to change the attribution back to Google, the chamber added.

The case, 26 O 869/26, orders Google to stop the specific false claims, with penalties for non-compliance, and assigns 80 percent of the legal costs to Google, with the two plaintiffs splitting the rest. Google said on June 12, 2026 that it ‘disagree[s] with the ruling and plan[s] to appeal,’ and that the company ‘invests deeply in the quality of AI Overviews to ensure that the overwhelming majority of responses provide accurate information.’

What the AI Said About the Two Munich Publishers

The AI Overviews in question did not hedge. In the affected queries, the summaries opened with affirmative claims such as ‘Yes, [company] is known for dubious business practices and is often perceived as a scam,’ then listed red flags, subscription-trap warnings, and tips for users. None of the underlying source pages made those claims about the plaintiffs, the court found, and the AI had blended information about other, genuinely questionable companies into the same answer. ‘The defendant’s own statements,’ the ruling called them, since only the Google algorithm produced the wording, the structure, and the conclusion. The plaintiffs sent Google a cease-and-desist letter earlier in 2026 asking for the false claims to be removed; the court found that Google’s response fell short.

The two plaintiffs are Munich-based publishers, and the court has not publicly named them, citing privacy considerations under German press law. Lausen Rechtsanwälte, a firm that has built a practice around AI-output disputes, argued the case for the publishers. The order requires Google to remove the specific objected content and ensure that the AI does not regenerate comparable false claims about the two companies.

The Federal Court of Justice Shield No Longer Fits

The Munich chamber spent most of its reasoning separating AI Overviews from the search-engine model that Germany’s Federal Court of Justice, the BGH, has shielded for years. Existing BGH case law treats search operators as indirect infringers: they make third-party content findable but do not adopt it, and a proactive duty to check would threaten how search works.

That reasoning does not transfer, the court ruled, because AI Overviews do not point. They generate ‘independent, new, and substantive statements’ by evaluating and combining content from multiple third-party sites, a step the chamber treated as a fresh act of speech. Only the operator, in this case Google, can check the result by comparing what its AI produced against the underlying web pages, the court added. A notice-and-takedown system built for links does not address that gap, since the offending text never appeared on a third-party page in the first place.

The chamber also pointed out that AI Overviews are ‘by no means absolutely necessary’ for using the internet, since traditional search results already help users sort through information. That detail matters because the BGH’s deference to search operators is partly practical: a duty to vet every result would break the product. No comparable practical justification exists for the AI summary, the court concluded, an extra layer on top of a working tool. The global implications of the Munich injunction are now being read into that single paragraph. The decision sits at the end of a short but deliberate legal roadmap:

• Earlier BGH rulings protected search-engine operators and autocomplete from direct liability for third-party content they merely listed.
• September 2025: the Frankfurt am Main Regional Court ruled, in case 2-06 O 271/25, that liability for AI Overviews is not per se excluded, though it stopped short of granting the specific injunction sought.
• May 28, 2026: the Munich Regional Court I, in case 26 O 869/26, prohibited specific false claims and assigned 80 percent of legal costs to Google.
• June 12, 2026: Google said it will appeal the Munich ruling, calling it focused on ‘specific and narrow errors’ rather than the foundational way AI Overviews work.

Google also cannot fall back on host-provider protections under the EU’s Digital Services Act, the chamber said, because the false content is Google’s own algorithmic output rather than a third-party upload. The Frankfurt decision in September 2025 had opened that door, and Munich has now walked through it.

Why Google’s ‘Users Can Verify’ Defense Failed

Google’s strongest argument at the hearing was that users can verify the AI’s claims themselves by clicking through to the linked sources, and that ordinary users already know AI-generated content should not be blindly trusted. That defense did not hold up. The AI overview is a ‘self-contained statement with independently understandable content,’ the chamber wrote, so the possibility of further research does not exempt the speaker from liability for the statement itself. Studies cited in the ruling show that ordinary users almost never click on the sources AI Overviews display, a point the court treated as evidence that the summary stands on its own.

An AI’s opinion is ‘not the expression of an acquired conviction of the persons expressing it, but the result of an algorithm,’ the court wrote.

The court drew a parallel to press law, where publishers are liable for teasers that read clearly on their own, even if readers never reach the full article. A protection gap was part of the reasoning: if Google were only liable for obvious violations, victims of false AI claims would have no real legal recourse, since the third-party sites whose content fed the AI had not made the statements in question. The false outputs were ‘primarily an expression of the defendant’s commercial activity,’ the ruling said, weighing Google’s interest in offering AI Overviews below the publishers’ privacy rights.

91% Accuracy Still Leaves Millions of Wrong Answers

The 91 percent accuracy rate for Google’s Gemini 3 AI Overviews, the model currently serving European users, sounds reassuring on its own. A startup analysis published in the New York Times put the figure at 91 percent factually correct, up from 85 percent under the previous Gemini 2 model, per the in-depth breakdown of the May 28 ruling.

That 9 percent gap is what made the Munich case possible. At Google’s scale, 9 percent of AI Overview answers translates into millions of false or unsupportable statements per hour, a volume the New York Post sized as ‘millions of false answers per hour’ after the same analysis. The court also had a separate problem: only 56 percent of the correct answers were backed up by the sources Google linked to them, according to the same study. The chamber did not cite the study by name, but its description of statements ‘that are not even made in the search results’ maps directly onto that 56 percent.

Click-through rates to the underlying publishers have moved in the same direction. Studies of AI Overviews in the wild show that pages cited inside an AI summary receive nearly 50 percent fewer clicks than they would from a standard search listing. That drop is the economic counterpart of the legal finding the Munich court made, and it tells publishers the same thing the court told Google: the AI has become the product, and the source pages have become the input.

• 91 percent: Gemini 3 AI Overviews answered correctly in the Oumi analysis for the New York Times, up from 85 percent under Gemini 2.
• 56 percent: of correct Gemini 3 answers, 56 percent could not be backed up by the sources Google linked.
• Nearly 50 percent: estimated drop in click-through rate to original sources when AI Overviews appear, per gHacks-cited studies.
• 80 percent: share of the legal costs the Munich court assigned to Google; the two plaintiffs split the rest.

Britain Forced a Separate Concession Earlier This Month

On June 3, 2026, two weeks before the Munich court published its reasoning, the UK’s Competition and Markets Authority ordered Google to let publishers opt out of having their content used in AI Overviews and AI Mode. The mechanism lives in Google’s Search Console, where website owners can now flag their pages as off-limits to the generative features. The CMA framed the move as a way to put publishers ‘in a stronger position to negotiate content deals with Google.’

The UK lever is competition law, not defamation, and the remedy is contractual control rather than damages. For Munich, where the question is who is liable for what the AI says, the answer is the operator. The two answers point the same direction, though: AI Overviews are no longer being treated as a passive wrapper around the open web, in London because of traffic, in Munich because of speech. Google told reporters it is testing the Search Console opt-out tools, and the company has said the broader CMA framework could serve as a blueprint for the EU market.

The Reasoning Reaches Beyond Google

The reasoning reaches well beyond Google. ChatGPT, Claude, and Perplexity all paraphrase web content into a single generated answer, the same shape the Munich court used to assign liability to the operator. None of them have been sued on this theory yet in Germany, but the legal test the chamber wrote could fit any AI system that produces ‘independent, new, and substantive statements’ from third-party sources.

The test, as the Munich court stated it, is whether only the operator can check what its AI produced against the underlying sources. For Google, that test is satisfied by a direct comparison between the AI’s output and the pages the model read; for an open-source model deployed by a third party, the analysis is more complicated, since the operator and the model developer may not be the same entity. What the ruling forecloses is the argument that the AI is just a more elaborate autocomplete, the position search engines have used for two decades to stay on the safe side of defamation law. The two existing models now sit on either side of a line the court drew:

Google’s appeal will test whether the higher court narrows that test, but the company’s public framing suggests a narrower fight. A spokesperson said the case ‘focuses on specific and narrow errors, not the foundational way AI Overviews displays web content,’ a position that concedes the underlying classification while disputing the remedy. Alphabet’s $80 billion raise for AI infrastructure shows how much the company is betting that the product survives whatever the courts do to it.

Frequently Asked Questions

Is the Munich ruling final?

No. The 26th civil chamber issued the order as a preliminary injunction in a summary proceeding. Google said on June 12, 2026 that it ‘disagree[s] with the ruling and plan[s] to appeal.’

What did the AI Overview say about the two Munich publishers?

The summaries opened with affirmative claims that the companies were tied to fraud, subscription traps, and dubious business practices. The court found that ‘none of the linked sources drew any connection between the plaintiffs and the shady companies the AI mentioned.’

Does the ruling change how AI Overviews work in Germany today?

For now, the injunction bars specific false claims about the two plaintiffs, with penalties for non-compliance. The broader legal question, whether the same reasoning applies to other defamation claims and other AI products, is unsettled until a higher court reviews the case.

Could the Munich reasoning apply to ChatGPT, Claude, or Perplexity?

No case has tested it yet. The legal test the chamber wrote could fit any AI system that produces ‘independent, new, and substantive statements’ from third-party sources, but the ruling applies to Google until another defendant forces a court to extend it.