A Munich district court recently made a landmark ruling, holding that the "AI Overview" displayed by Google in search results is Google's own content and not simply an aggregation of search results. Therefore, Google must bear direct infringement liability for the false statements contained therein.
In this case, Google’s AI overview falsely linked two Munich-based publishing companies to scams, subscription traps and unfair business practices under specific search queries. The court found that the AI system mixed information about other questionable businesses with the plaintiffs and fabricated associations that did not exist in any of the linked sources. After the plaintiff issued a lawyer's letter requesting to stop the infringement, Google did not respond fully, so the court issued a temporary injunction prohibiting the continued dissemination of relevant false accusations through AI Overview (Case No. 26 O 869/26).
The court emphasized that the AI overview was not a “search result” in the traditional sense. Compared with regular searches that only list external links, Google's AI overview will rewrite and evaluate the retrieved content "in its own language and according to its own structure." On the page involved in the case, the AI overview even begins with an affirmative sentence such as "Yes, [company] is known for questionable business practices" and organizes sections including "summary", "suspicious signals" and "anti-fraud suggestions" to form an independent and self-consistent narrative. The court pointed out that the AI overview not only reproduced existing information, but also made new statements that were "completely not found in any search results" and that these contents belonged to Google's own statements. Because Google develops and provides this AI feature to users and has exclusive control over its algorithm, it must be directly responsible for the content generated.
In terms of liability determination, the court clearly distinguished the legal status of traditional search engines and AI overview. The existing jurisprudence of the German Federal Court of Justice (BGH) has given limited liability to search engines and auto-complete functions on the grounds that they only allow third-party content to be retrieved and are only liable for indirect infringement in principle. If they are required to fully review the results in advance, it will threaten the viability of the search service. However, the Munich court held that this logic does not apply to the AI overview. Traditional searches only point to external websites, while AI Overview generates "independent, new substantive statements" by evaluating and combining the content of multiple websites. These statements can only be systematically verified by Google itself, at least by comparing the content of the cited websites with the AI output. The court also emphasized that AI overview is "by no means a necessary function for using the Internet". Ordinary search results are enough to help users filter information. AI overview is just an additional service, so its responsibility cannot be relaxed on the grounds that "the technology is indispensable".
During the trial, Google argued that users can click on the source link below the AI overview to verify the authenticity of the content, and that “users generally know that they should not blindly trust the information generated by AI.” This statement contrasts sharply with the business reality of Google’s massive promotion of AI overview capabilities, while also ignoring the fact that AI summaries often lack a clear and traceable relationship to the original source. The court rejected this defense, stating that the fact that a statement could be disproven through further search did not "ordinarily relieve the publisher of the statement from liability." In this case, the AI Overview constituted a "complete statement that is understandable in itself" and did not alert the user that its content might be ambiguous or unreliable. Research shows that the proportion of users clicking source links in the Google AI overview is extremely low, which confirms the court’s judgment on user behavior patterns.
The Court further invoked an analogy from the field of press law: the media is responsible for the content of independently understandable news headlines or "clickbait" headlines, even if the reader never reads the full text. If Google's "user self-examination" argument is accepted, the actual effectiveness of the AI overview will be greatly weakened, because it means that its content is "generally regarded as unreliable", which is contrary to Google's original intention of promoting this feature. In addition, if only “obvious” illegal content is held accountable, victims will have almost no way to defend their rights when faced with false accusations fabricated by AI: the third-party website that provided the underlying content has never published these remarks, and the victims can neither sue the source nor effectively hold Google accountable according to traditional search rules. Therefore, the court found that Google could not invoke the exemption clause in the Digital Services Act that applies to hosting service providers, nor could it simply apply the search engine "notice-and-takedown" process to avoid liability.
At the level of freedom of speech, the court also made exemplary discussions. The ruling pointed out that the "views" generated by AI are not expressions of subjective beliefs formed by natural persons based on their own cognition, but are the product of algorithmic calculations. Google's provision of AI-assisted search services is "first and foremost a reflection of its business activities" and can at best be seen as a secondary reflection of the company's interest in freely expressing its views and positions. When weighing the plaintiff's reputation against Google's interests, Google's interests should be secondary, especially when the relevant statements are based on factual misrepresentations and wrongly link the plaintiff to certain businesses.
Regarding the specific verdict, the court supported the plaintiff on most of its claims. The ruling expressly prohibits Google from spreading a series of claims through AI overview, including accusations of fraud, association with questionable companies, subscription traps, phone calls that never occurred, unavailability of contact or non-availability of service, and only two minor requests were rejected. Although the specific copy involved in the case has been taken offline, the court believed that the risk of repeated infringement still exists because Google has not made a formal statement to stop infringement with penalty clauses, and its algorithm may also generate similar content again under similar queries. In terms of litigation costs, Google needs to bear 80% of the costs, and the two plaintiffs each bear 10%. The court also noted that the judgment in principle has potential cross-border implications.
The impact of the ruling extends far beyond the parties in this case. According to an analysis conducted by AI startup Oumi for the New York Times, about 91% of the answers in Google AI overview tests based on the current Gemini 3 model were judged to be correct. This accuracy rate may seem "reliable enough" for daily use by the average user, but at Google's scale, even a 9% error rate means that millions of incorrect answers may be generated every hour. If a significant number of these errors involve defamation or false accusations against companies or individuals, this will constitute a serious legal risk not only for Google, but also for other similar service providers (including ChatGPT, Claude, Perplexity, etc.). Oumi's analysis also shows that even among the samples that were deemed to have "correct answers", about 56% of the answers could not find direct support in the link sources displayed by Google, making it difficult for users to trace the basis for these conclusions.
The Munich Court responded to this risk of "disconnection from the source": when the statements generated by the AI system do not exist in any linked source, they already constitute independent claims, and the operator must be responsible for these claims. It is not yet known whether this logic can be maintained on appeal and whether other jurisdictions will adopt similar ideas. Google has not publicly commented on the ruling in this case. However, if the judgment is reverberated internationally, the impact may not be limited to Google, but all AI service providers whose selling point is “automatically extracting, rewriting, and summarizing from online content.”