The Regional Court of Munich issued a preliminary ruling against Google on May 28, 2026, finding the company directly liable for false and defamatory statements generated by its AI Overviews feature, and requiring Google to prevent the dissemination of erroneous or inaccurate claims through its search engine. You can see a copy of the Regional Court of Munich’s preliminary ruling here; for an English translation of the ruling (created by DeepL) click here.

The case involved two Munich-based publishers who alleged that their companies’ reputations were damaged when Google’s AI Overviews incorrectly linked them to scams, subscription traps, and fraudulent business practices. The court found those claims were not present in any of the underlying source material used to generate the AI Overviews summary. The affected companies sent Google a cease-and-desist letter, and in response, Google denied liability, noting that its automatic summary feature warns users that the information may contain errors and should be independently verified.

Court: AI Overviews aren’t the same as search results

The court’s central finding rests on a legally significant distinction: AI Overviews are not search results. Traditional search engines have historically enjoyed a degree of legal protection because they direct users to content created by third parties. German courts had previously reasoned that imposing a proactive duty to verify every indexed result would render search engines operationally unviable. The Munich court found that this reasoning does not apply to AI-generated summaries. When Google’s AI produces a statement such as “Yes, this company is known for dubious business practices,” it is not retrieving a third-party opinion; it is generating one. The court described these outputs as “independent, new, and substantive statements,” classifying Google as a direct infringer rather than a passive intermediary.

Google’s primary defense, that users should independently verify AI-generated content, was rejected. The court drew a parallel to German press law, noting that publishers bear liability for headlines and teasers that function as standalone statements, regardless of whether readers engage with the full article. AI Overviews operate in the same manner. A July 2025 Pew Research Center study found that Google users clicked a link inside an AI summary in just 1% of visits to pages that showed one, so the AI’s output functions as the definitive response for most readers. The court found that advising users to verify information while simultaneously positioning AI Overviews as an efficient research tool is inconsistent and does not constitute a valid legal defense.

The ruling also addresses questions that are significant for the broader AI industry and may resonate in the United States. The court held that AI-generated content does not warrant the same free speech protections afforded to human expression because algorithmic output reflects commercial and operational logic rather than individual conviction. This distinction is directly relevant to how future German and European courts may balance the interests of AI operators against those of individuals and organizations harmed by inaccurate machine-generated content.

The court further determined that Google could not rely on the European Union’s Digital Services Act host-provider protections, nor could it fall back on standard notice-and-takedown procedures applicable to conventional search engines. For companies including OpenAI, Anthropic, and Perplexity, whose systems similarly synthesize and paraphrase web content, standard disclaimer language in terms of service agreements may not provide sufficient legal protection. The court argued that without a cause of action against the AI models, victims of false statements in AI Overviews would be legally powerless, since the original sources never made those statements and, therefore, could not be subject to legal action.

Google’s AI Overviews are only 91% accurate, NYT finds

The Munich court’s findings are supported by a New York Times article, How Accurate Are Google’s A.I. Overviews?, which commissioned a study by AI start-up Oumi using the SimpleQA benchmark across 4,326 Google searches. The study found that AI Overviews were accurate 85% of the time as tested in October 2025 (Gemini 2) and 91% of the time as tested in February 2026 (Gemini 3). With over five trillion searches annually (according to the Times story), inaccurate responses number in the tens of millions per hour.

The Times study also found that more than half of the accurate responses were “ungrounded,” with source support often lacking, and that this figure increased from 37% to 56% with Gemini 3. Additional issues included AI Overviews citing sources like Facebook posts and Reddit threads, and vulnerability to manipulation, where a single blog post could cause Google to present false information. The study also found that AI Overviews can give conflicting answers in seconds and may produce false responses even when sources are correct.

If sustained on appeal and adopted in other jurisdictions, the Munich court’s reasoning would establish that operators of AI systems bear direct responsibility for the content those systems produce. The decision also raises the possibility that the same theory of liability could apply to corporate chatbots that provide false information.