Google has spent years trying to convince users that AI Overviews are a helpful layer on top of search: a neat little answer box that digests the web and saves everyone a few clicks. A Munich court looked at the same feature and saw something legally different from a search result. If Google's AI writes a new summary that falsely connects a company to scams, the court said, Google may own that statement.

The dispute involved two Munich publishers. According to coverage by The Decoder, WIRED, and Search Engine Land, Google's AI Overviews falsely associated the publishers with scams, subscription traps, questionable business practices, and related accusations. The linked source material did not make those connections. The AI had apparently blended information about other suspect companies with information about the publishers, then produced a confident answer that looked complete enough to stand on its own.

That is the central problem with AI search. Traditional search results usually point somewhere. They quote or summarize snippets from pages, but the user can see that the statement comes from a third party. AI Overviews synthesize. They rewrite, combine, rank, infer, and present the output as a finished answer. When that answer is wrong, the victim may not have an original publisher to sue, because no original publisher said the thing. The machine assembled it.

Search result or publication?

Google's defense leaned on the familiar search-engine model. Search providers generally get more protection when they index and display third-party content. If a website says something false, the website is the primary speaker. The search engine is more like a pointer.

The Munich court did not accept that framing for AI Overviews. The reports describe the court treating the summaries as Google's own generated content. The AI did not merely list pages. It produced independent statements, in its own structure, that allegedly went beyond the linked material. That distinction is the whole case. If the feature generates an accusation that no source actually contains, calling it "search" stops being a useful description and starts looking like a liability escape hatch.

Google also argued that AI Overviews warn users to verify information. That argument has a certain gallows charm. "Our product may be wrong, please check it" is not much comfort to a small publisher suddenly labeled scam-adjacent in the answer box at the top of the world's dominant search engine. A disclaimer does not unpublish the claim. It just adds a footnote to the damage.

The court reportedly ordered Google to stop repeating most of the challenged statements and assigned Google 80 percent of the legal costs, with the publishers paying the rest. The ruling was temporary and could still be challenged. Google told The Decoder that AI Overviews are designed to reflect information on the web and that the company was reviewing the decision.

Why this belongs here

Vibe Graveyard already has several AI Overview stories because the feature has produced a varied buffet of nonsense: glue pizza, rock eating, fake medical summaries, and defamation complaints. This Munich ruling adds a different kind of consequence. It is not just "the answer box got something wrong." It is a court saying the operator of the AI system can be treated as directly responsible for what the generated answer says.

That matters because AI search changes the harm model. A normal bad search result may send users to an inaccurate webpage. A bad AI Overview can put the inaccurate statement directly in Google's voice, above the links, in the spot users are trained to treat as the answer. For a business, publisher, doctor, artist, or local service provider, that placement can be reputationally expensive before anyone has time to send a cease-and-desist letter.

The case also exposes the problem with retrieval as a magical liability solvent. AI Overviews are grounded in web sources, in theory. In practice, grounding is not the same as truth. A model can cite pages that do not support the generated statement. It can conflate similarly named entities. It can pull a negative fact from one company and attach it to another. It can synthesize a crisp accusation from a pile of fragments that never belonged together. Once the system writes the output, the operator cannot simply point at the web and say the web made it do that.

The product design failure

The design temptation is obvious. Search wants to answer, not just route. AI makes answer generation feel cheap and scalable. But when the product shifts from indexing statements to creating statements, the product also inherits the risks of creation: defamation, false light, business disparagement, consumer confusion, and plain old reputational wreckage.

This is especially risky for queries about companies and people. A generated answer about business conduct is not like a generated recipe. False claims about scams, fraud, safety, criminality, credentials, or professional misconduct can damage trust immediately. The answer box does not need to be malicious to be harmful. It only needs to be wrong in a place people see.

The fix cannot be "users should verify." That is backwards. The system publishing the answer should verify before it displays the answer, especially when the output makes negative claims about identifiable entities. At minimum, AI search needs stronger entity disambiguation, claim-source matching, defamation-sensitive guardrails, and a fast removal process that does not require victims to litigate their way out of an invented summary.

The Munich ruling may not be the final word, but it draws a useful line. If an AI system writes a new factual statement, and that statement harms someone because it is false, courts may not treat it like a neutral list of blue links. Search engines wanted answer engines. Answer engines come with answer-shaped responsibility.