A German court in Munich has issued an injunction against Google, ordering it to stop publishing false claims about two local publishers that its AI Overviews had wrongly connected to fraud and shady business practices. The AI had mixed up these publishers with genuinely problematic companies, inventing connections that appeared in none of the linked sources. Google was told to cover 80% of the legal costs.
The legal logic at the heart of this ruling is worth understanding clearly. For decades, search engines have been shielded from liability because they simply point users toward content that other people wrote. Germany's higher courts had previously ruled that holding search engines fully responsible for everything they link to would break how search works. The Munich court found this reasoning does not apply to AI Overviews.
AI Overviews do not just point. They write. The court said they generate what it called "independent, new, and substantive statements" by pulling from multiple sources and combining them into new text. That makes Google the author of those statements, not just a directory. And only Google can check whether those statements are actually true, the court added.
Google argued at the hearing that users could always click the linked sources to verify the AI's summary. The court was not impressed. Studies show that users who see an AI Overview are far less likely to click through to source websites. In one study of 900 users, those shown an AI summary clicked on traditional search results only 8% of the time, compared to 15% for users who did not see a summary. The court also drew a parallel to press law: a newspaper teaser is legally its own statement even if readers never read the full article behind it.
The accuracy numbers add important context. An independent analysis commissioned by The New York Times tested over four thousand Google searches and found AI Overviews were correct about 91% of the time with Google's current model. That sounds solid until you apply it to Google's actual volume. The platform processes more than five trillion searches a year. A 9% error rate at that scale produces tens of millions of wrong answers every hour.
The sourcing problem may be even more telling. The same analysis found that 56% of correct answers could not be backed up by the sources Google linked. The AI arrived at the right answer, but the pages it cited as evidence did not actually support it. That makes it nearly impossible for anyone to verify what the AI is telling them, even when it happens to be right.
This ruling is still a preliminary injunction from a regional court. Google can appeal, and this is not binding across Germany's legal system. But the court explicitly said its reasoning could have international reach. The same logic, if it holds, applies to every AI-powered answer engine: ChatGPT Search, Perplexity, Microsoft Copilot, and others. All of them write their own summaries rather than simply listing links.
For businesses, especially any company with a public reputation to protect, this matters in a direct way. If an AI summary incorrectly ties your company to bad practices, a legal route to force a correction now exists in at least one major jurisdiction. If you run a business that uses AI-generated content in a customer-facing context, this case signals that courts are starting to treat that content as your own speech, not a neutral tool's output.
The broader lesson here is that the "AI can make mistakes" disclaimer that tech companies have relied on is looking increasingly thin as a legal shield. The Munich court's position is essentially that you cannot deploy a confident-sounding system to two billion monthly users and then tell the victims of its errors that they should have known better than to trust it.