TL;DR
- A Munich court ruled Google is directly liable for false statements generated by its AI Overviews
- The AI smeared a real publisher, Verlagshaus24, and its GeraMond imprint as scammers, inventing connections that appeared in no source
- The court called the overviews "Google's own words," not search results, so the search-engine liability shield no longer applies
- Google's AI has also told people to put glue on pizza, eat rocks, and cook spaghetti with gasoline
- Even 91% accuracy means millions of wrong answers each hour, and you can't edit what Google says about you
Verlagshaus24 is a Munich publisher with twelve imprints and an online shop. GeraMond is its subsidiary, putting out books and magazines about technology and history. Real businesses, with real reputations.
Then one day in January, a customer searched for the company name followed by the phrase "fraud scheme." It is a phrase Google's own autocomplete suggests. At the top of the page sat an "AI Overview" that answered, in the confident tone of a consumer-advice article: "Yes, V.24 is known for dubious business practices. Yes, there are indications of fraudulent schemes and dubious practices."
The AI built a whole structure around that claim. It listed the "Characteristics of the Alleged Fraud Scheme." It warned about subscription traps and invoices for "premium gold packages." It told readers the company changed names and URLs to avoid detection, and that it was in contact with a genuinely shady SEO firm. Then it offered a box titled "What You Can Do."
None of it was true. The only source the overview cited was an article about a completely different set of companies. The publishers appeared nowhere in it.
What the publisher did about it
Their lawyer sent Google a cease-and-desist letter on February 2, 2026. Google replied that it could not process the complaint at that address and pointed to an online form. They resubmitted through the form. On February 10, a fresh search still returned a near-identical false summary. Google kept asking for more information.
So the publishers went to court. The Regional Court of Munich heard the case on April 23 and granted a preliminary injunction on May 28.
Why the court sided with the publisher
The court's reasoning matters more than the ruling itself, and Google is appealing, so the ruling is not final. But the logic is a milestone.
For decades, search engines enjoyed a shield. A search engine does not create the pages it lists, it points to them, so the law treated it as an intermediary, not an author. German courts had protected Google's autocomplete and search listings on that basis.
The Munich court said AI Overviews are different. A search result links to a website. An AI Overview writes its own text: it rewrites information in its own words, adds its own structure, and invents claims that appear nowhere in the linked sources. The court called these "independent, new, and substantive statements" and "Google's own words," because only Google controls the algorithms that produced them.
Google argued that users could check the sources themselves. The court rejected that too. An AI answer reads as a self-contained statement, with no hint that it might be wrong, and studies show almost no one clicks the source links anyway. The court even noted that an AI opinion is "not the expression of an acquired conviction," just the output of an algorithm, so it deserves less free-speech protection.
The court ordered Google to stop spreading the false claims and to pay 80% of the legal costs.
The greatest hits: what Google's AI has told people
If you think the publisher's case is a fluke, the public record says otherwise. AI Overviews have a well-documented history of confident nonsense.
Ask how to make cheese stick to pizza, and Google's AI suggested adding "non-toxic glue" to the sauce. The source was an eleven-year-old joke from a Reddit comment. It has also told users that "geologists recommend humans eat one rock per day," drawn from a satirical article on The Onion.
A reporter asked whether gasoline could cook spaghetti faster. The answer: "No, but you can use gasoline to make a spicy spaghetti dish," complete with a recipe. And as recently as 2026, a user who set up a projector to watch a sunset was told by Google's AI that the sunset had already happened.
These are funny until you remember how many people take the answer at face value.
The math that should worry you
Here is where it stops being funny. An analysis by AI startup Oumi found that Google's AI Overviews answer correctly about 91% of the time. That sounds great, until you multiply the remaining 9% by the number of searches Google serves every hour. Even a tiny error rate at that scale means millions of wrong answers every hour.
Worse, the same analysis found that 56% of the answers Google's AI got right could not be backed up by the sources it linked. The machine produces claims whose origins nobody can trace. That is exactly the problem the Munich court identified: the AI generates its own version of reality, loosely based on the web.
When it is your name in the summary
Your reputation no longer lives only in reviews, articles, and your own website. It also lives in AI summaries you do not control and cannot edit.
A customer searches "is [your company] legit" before buying. Google's AI writes the answer. If that answer is wrong, you have no way to correct it directly. You can update a Google Business Profile, but nobody logs into Google and edits an AI summary. The court just made Google legally responsible, but being right on paper is cold comfort when the customer has already walked away.
The same logic applies in Moldova, whatever language your customers search in. The AI answer is generated, not indexed, and it works the same way everywhere.
How to protect yourself
Treat an AI summary as a hint, not a verdict. A few habits keep you safe.
Go to the source. If an AI answer matters to a decision, open the links it lists and read the actual page. If the links are missing, or the claims are not there, treat the answer as unverified.
Search the exact claim. Type the precise sentence you read. A false claim rarely survives contact with several independent sources.
Check official pages, not summaries. For prices, hours, legal requirements, or contact details, go to the official website or the authority that issues the information. One look at the source beats ten AI answers.
Watch your own listing. Search your business name in different languages and note what the AI says. If something is wrong, that is your early warning. Report the issue, and make sure your own website states the facts clearly enough to be picked up.
"Even a tiny error rate at Google's scale means millions of wrong answers every hour."
The Munich ruling settles something that was up for grabs: a company is responsible for what its machine says, even when the machine is only "helpfully" summarizing. Until the appeal lands, the safest approach is boring. Verify before you trust, and make sure your business says about itself, clearly, what you want customers to read.
ShiftIT is a custom software and AI automation agency based in Chișinău, Moldova. We help businesses automate operations and scale their digital presence.