Landmark Ruling: Google Can Be Held Responsible for What Its AI Writes

Are artificial intelligence companies responsible for what their chatbots say? A regional court in Munich answered that question unequivocally this week, ruling that Google is directly responsible for the content generated by its AI Overviews, even when they are based on information from other websites. The decision is a precedent that could affect not only Google but the entire AI industry, including services such as ChatGPT, Claude and others that generate answers by combining and paraphrasing information from the web.

The ruling was issued in the context of a request for a temporary injunction after two publishing companies from Munich argued that Google’s AI summaries portrayed them as companies linked to fraud, subscription traps and questionable business practices, despite, they said, there being no factual basis for this. According to the court, the AI system mixed information about other companies, which had in fact previously been associated with problematic conduct, with the plaintiffs, and even created connections that did not appear in any of the sources to which it linked. The companies sent Google a warning letter demanding that the publications be removed and that it commit not to repeat them, but according to them the company did not provide a sufficient response.

Not a search engine, but a content creator

One of the most important findings in the ruling is that AI summaries are not similar to the results of a regular search engine. While a traditional search engine merely directs the user to other websites, Google’s AI system rewrites the information, builds its own structure and presents its own conclusions. In the case at issue, for example, the summary opened with a clear assertion that one of the plaintiff companies was “known for questionable business conduct,” and later included a summary, a list of “red flags” and tips for users on how to be cautious around it.

According to the court, these are not search results but independent statements by Google itself. The judges noted that some of the claims that appeared in the summary did not appear at all in the sources to which Google linked. None of the websites connected the plaintiffs to those questionable companies, and the link between them was created solely by the algorithm. That is also why the court defined the summary as “content of the defendant itself.” Google developed the AI system, operates it and offers it to the public, so it is also responsible for what it generates, since only it has control over the algorithms and how they operate.

AI summaries are not an essential service

The court also reviewed earlier rulings by Germany’s Federal Court, which had previously held that search engines and autocomplete services benefit from limited liability because they merely make third-party information accessible. But according to the judges, that logic does not apply to AI systems. In their view, a regular search engine only shows links to external websites, while, as noted, an AI system generates “new, independent and substantive statements” by combining information from various sources. Therefore, only Google is able to check whether these claims are true, by comparing the original sources with the text the system created. The judges also noted that AI summaries are not an essential service for using the internet. Regular search results already provide users with the information they need, and the summaries are merely an addition intended to make reading and receiving information easier for users.

During the hearing, Google argued that users can open the sources to which the AI summary refers and verify for themselves whether the information is correct. The company also claimed that the public now understands that information generated by artificial intelligence should not be trusted blindly. The court rejected that argument. In its view, the mere ability to independently check information does not absolve a publisher of responsibility for publishing false information. The AI summary stands on its own, is written clearly and presents an independent message that does not include disclaimers or warnings about its reliability. The judges even cited studies showing that most users do not click the links attached to AI summaries at all, but instead rely on the answer that appears at the top of the page. They also argued that if the public were to treat AI summaries as unreliable from the outset, it would significantly reduce the usefulness of the service.

A ruling with international implications

According to the ruling, this is the result of an algorithm, not an expression of a position, belief or worldview. The judges held that operating an AI service is прежде всего a business activity of Google, and therefore when it conflicts with the rights of people harmed by the publication of false information, the company’s interest is pushed to the bottom of the legal hierarchy. Accordingly, the court prohibited Google from publishing claims linking companies to fraud, questionable companies, subscription traps, phone calls that never took place or business unavailability.

It was also determined that even if the specific texts are no longer displayed, there is still a real risk of recurrence, since Google has not committed to refraining from publishing similar material in the future, and there is nothing preventing the algorithms from generating the same claims again.

The implications could be particularly broad in light of a recent study conducted by the AI company Oumi for The New York Times, which found that Google’s AI summaries based on the Gemini 3 model provide correct answers in about 91% of cases, a high rate of accuracy, but at Google’s scale that still amounts to millions of incorrect answers every hour. The study also found another troubling figure, 56% of the correct answers are not supported by the sources to which Google links, so users cannot trace the origin of the information.

That is precisely the issue the Munich court sought to resolve, when an AI system generates new claims that do not appear in any linked source, responsibility for them lies with the company operating the system. The ruling may still change on appeal, and Google has not yet responded to the decision. But if the principle established in Munich is adopted in other countries as well, it could mark a turning point for the entire artificial intelligence industry and require other companies too to take full legal responsibility for what their models write.