Two Munich publishers discovered one day that Google was linking them to fraud and dubious business practices. Not in some obscure discussion, but right at the top of the search results, in the so-called AI Overviews. These are the AI-generated summaries displayed above the regular links. Yet none of the articles referenced by the summaries made any such claim.
According to the court, Google’s AI took information about other companies that had previously been identified as possible fraudsters and combined it with data about the two publishers. This created a connection that did not appear in any source. The artificial intelligence simply made it up. The publishers first tried a more restrained approach. They sent Google a formal notice demanding that it stop. But Google failed to prevent similar false statements, and that inaction ultimately came back to haunt it. Both websites took the matter to court.
The Munich Regional Court issued a preliminary injunction. The ruling is clear: Google is liable for what its AI summaries claim.
AI summaries vs. traditional search
Germany’s Federal Court of Justice previously ruled that companies such as Google have only limited liability for ordinary search results. A search engine is merely a signpost pointing to third-party content. When it ranks links, that does not mean it guarantees their content. It is a channel, not a publisher. But both logically and according to the court, AI Overviews are in a completely different league.
The judges concluded that artificial intelligence does not summarize content passively. It reformulates and combines it, and sometimes even invents facts that appear nowhere in the sources. It then serves all of this at the very top of the results. The court therefore refers to “independent, new, and factual claims.” In other words, these are no longer someone else’s words, but words attributable directly to Google. And why Google in particular? Because only Google can intervene in the models and logic that generate those claims. Only Google can reliably stop the system from repeating the same lie again and again.
The excuse that failed
Google defended itself in court by arguing that most users know they should not blindly trust information from AI. Moreover, the summaries contain links to sources, so anyone can verify the information independently. The court rejected this defense, arguing that the ability to verify a claim generally does not absolve you of responsibility. If it did, victims of false statements would be left with virtually no recourse. The original sources made no such claims, so they could not be sued. And the defamed person or company would have no one to hold accountable.
Imagine it happening to you. If someone writes something outrageously false about you, the fact that it can easily be checked elsewhere does not let them off the hook for the lie.
The court also addressed one sensitive issue. Can a statement generated by artificial intelligence be protected by freedom of expression? According to the court, no. An AI output is not an expression of a conviction formed by the person making the statement, but the product of an algorithm. According to the court, AI search is “primarily an expression of Google’s commercial activity” and only marginally an expression of an interest in freely sharing an opinion. This line of reasoning may well appear in other legal disputes around the world.
There are more cases like this
The figures suggest that this is not an isolated incident. Last year, Pew Research found that when an AI summary appears in search results, people click on regular links far less often. The rate fell from roughly fifteen percent to eight percent. Users receive a quick, confident answer and do not investigate any further.
An analysis by The New York Times also showed that AI Overviews provide the correct answer in roughly nine out of ten cases. That sounds good until you consider the scale. Google processes more than five trillion queries a year. Even a small error rate therefore translates into millions of incorrect answers.
And now for the worst part. According to the analysis, more than half of even the correct answers were “unsupported.” The websites cited by the AI did not fully substantiate its claims. So even when Google links to a page containing the correct information, its artificial intelligence can still produce a falsehood.
Google is not giving up
Google disagrees with the ruling and has announced that it will appeal. “This case concerns specific, narrowly defined errors, not the fundamental way in which AI Overviews display web content. We disagree with the decision and plan to appeal,” a company spokesperson said. Google also claims that the overwhelming majority of summaries are accurate, although it acknowledges that they may occasionally miss the context or misunderstand a website’s content. It says it acts quickly against violations of its policies.
In this case, however, the company must pay eighty percent of the legal costs and withdraw some of the statements that the court deemed defamatory.
The ruling could have repercussions far beyond Germany. Companies such as OpenAI, Anthropic, and Perplexity use the same warning that “AI can make mistakes.” And if the Munich court’s reasoning stands, that small disclaimer in their terms of use will not protect them in court.
Sources: malwarebytes.com, wired.com and tech.yahoo.com
