Share This Article
Google AI Overviews liability is now a reality, after the Munich Regional Court I held Google directly responsible for the false information its AI Overviews invent, in what is believed to be the first ruling of its kind on AI-generated search answers.
In its judgment of 28 May 2026 (case no. 26 O 869/26), the Landgericht München I classified Google not as a neutral intermediary, but as a direct infringer whose AI produces false statements as its own content. The reasoning could reshape liability for every provider of synthesized AI responses.
In this article, I explain what the court decided, why the reasoning matters far beyond this single case, and what businesses deploying generative AI should take away from it.
Google AI Overviews Liability: What the Munich Court Decided
The facts are straightforward, and they are exactly the kind of failure that keeps general counsels awake at night. Google’s AI Overviews falsely linked two Munich-based publishers to scams, subscription traps, and dubious business practices. The AI pulled in information about genuinely shady companies and attributed it to the plaintiffs, even though those connections appeared in none of the cited sources.
The plaintiffs were two companies forming part of a Munich media group spanning twelve branded imprints, including a subsidiary publishing under the GeraMond brand. They sent Google a cease-and-desist letter. When Google did not respond appropriately, they went to court, and on 28 May 2026 the 26th civil chamber granted a preliminary injunction barring Google from repeating the false statements.
What makes the decision significant is not the error itself. Hallucinations are well documented. It is the legal reasoning that strips away the protection search engines have relied on for years.
Why AI Overviews Are Not Search Results
The court drew a sharp line between traditional search and AI summaries, and this distinction is the heart of the Google AI Overviews liability finding.
Conventional search results merely present indexed third-party content with a title, a snippet, and a link. AI Overviews go further. According to the court, they generate a coherent, flowing text that evaluates multiple sources and combines them into an independent answer. To the average user, that answer looks like information coming directly from Google, rather than a forwarding of third-party content.
The chamber therefore concluded that these were Google’s “own statements”. Crucially, the AI made claims that did not even appear in the linked sources. As the court put it, Google built the AI, Google offered it to users, and Google alone controls the algorithms behind it, so Google must answer for what it produces.
The court also examined existing German Federal Court of Justice (BGH) case law, which gives traditional search engines and autocomplete only limited, indirect liability because they simply make third-party content findable. That reasoning, the chamber held, does not transfer to AI Overviews. A search engine points to external websites. AI Overviews instead generate independent, new, and substantive statements by evaluating and combining content, and only Google can verify those statements against the underlying sources. You can read the original reporting on heise online and a detailed analysis on The Decoder.
The “Users Can Check the Sources” Defense Failed
Google’s main argument was that users could simply click through and verify the sources themselves, and that people generally know AI-generated information should not be blindly trusted. The court rejected this defense outright.
First, the possibility of disproving a statement through further research does not normally exempt anyone from liability for that statement. Second, the user-behavior data cut against Google: research shows people click a source link inside an AI summary in just 1% of visits. The AI Overview was understandable on its own, contained a self-contained statement, and gave no signal of unreliability.
The court even drew a parallel to press law, where a publisher is liable for a teaser that stands on its own, even if readers never open the full article. As the chamber noted, Google’s own argument would also undermine the value of the feature, because an overview openly recognised as unreliable would be of little use to anyone.
Because the statements were the platform’s own content, Google could not fall back on host-provider protections under the Digital Services Act, nor on the standard notice-and-take-down process available to search engines.
Less Free Speech Protection for AI-Generated Statements
The court went one step further on free speech. An AI’s output, it reasoned, is not the expression of a personal conviction but the result of an algorithm. Offering AI-powered research is, above all, an expression of Google’s business activity, and only secondarily an exercise of free expression.
Weighing the plaintiffs’ personality rights against Google’s commercial interests, the court placed Google’s interests second, especially because the contested statements rested on untrue facts. The plaintiffs supported their case with sworn affidavits confirming they had no connection whatsoever to the companies the AI had linked them to. Google was ordered to cover 80% of the legal costs, with the plaintiffs bearing 10% each.
Why Google AI Overviews Liability Is a Governance Issue
The ruling goes well beyond a single dispute, and this is where Google AI Overviews liability becomes a board-level concern rather than a niche litigation point.
The accuracy numbers tell the story. An analysis carried out for the New York Times found that Google’s AI Overviews, running the current Gemini model, answered correctly around 91% of the time. That sounds reassuring, until you remember Google’s scale: even a small error rate translates into millions of wrong answers. The same analysis found that a majority of even the correct answers could not be traced back to the linked sources. In other words, the system regularly produces statements whose origin users cannot verify, which is precisely the problem the Munich court addressed.
For any business deploying generative AI, the practical lessons are clear:
- A disclaimer is not a shield. “AI-generated, may contain errors” did not protect Google, and it is unlikely to protect anyone else producing self-contained, factual-looking outputs.
- Defamation risk concentrates on real entities. Fabricated claims that a named company or individual is involved in fraud carry the highest exposure.
- Grounding and traceability matter legally, not just technically. If your model cannot tie its statements to verifiable sources, you own those statements.
- Notice-and-action processes need teeth. Failing to react properly to a cease-and-desist letter weighed against Google, and the absence of a binding undertaking left the risk of repetition intact.
This decision should also be read alongside the broader European liability trend. From 9 December 2026, the new framework on AI liability under the Defective Products Directive will treat software and AI systems as products subject to strict liability, while the wider debate on platform design liability continues to expand the responsibilities of major operators. The direction of travel in Europe is unmistakable.
A Few Important Caveats
Some qualifications are essential. This is a temporary injunction, not a final judgment, and Google has stated that its AI Overviews are designed to reflect information already available on the web and that it is reviewing a decision it considers not yet final. The reasoning may still be tested on appeal.
Yet if it is upheld, the consequences extend far beyond Google. The same logic would apply to any provider of synthesized AI answers, from ChatGPT to Claude to Perplexity, and to enterprises building retrieval-augmented chatbots on top of them. The protective shield that intermediaries have relied on does not stretch to systems that author their own answers.
One thing is certain: Google AI Overviews liability is no longer an abstract debate. It is a live risk that companies offering generative AI outputs must factor into their governance, their contracts, and their content controls today.
On a similar issue, you can read the article “AI Liability under the Defective Products Directive: Software and AI as Products from 2026“. Also, don’t miss our AI Law Journal available HERE.