When AI Lies, Who Pays? A German Court Just Pointed at Google

Type a question into a search box today and you don’t get links anymore. You get a finished answer. Convenient. But what happens when that answer is a lie? A German court just delivered a pointed reply: if Google’s AI made it up, then it’s still Google talking.

What Actually Happened

Google stamps a feature called AI Overviews at the top of its search results. You search, and instead of scanning ten blue links, you get an AI-written summary that stitches together several web pages into one tidy paragraph. It’s that gray box you’ve probably scrolled past a hundred times.

The problem is that this AI occasionally invents things that sound true but aren’t. The industry has a polite word for it: hallucination. The model confidently states facts that simply don’t exist.

That’s the heart of the German case. AI Overviews served up false information about a specific subject, and the party that says it was harmed took Google to court over it.

The One Thing the Court Zeroed In On

This ruling matters because it goes straight at Google’s favorite defense.

For years, Google’s line has been some version of: “We don’t create information, we just surface what’s already on the web.” A search engine, the argument goes, is a neutral intermediary — like a librarian who points you to a book but isn’t on the hook for what’s printed inside it. In the US, that logic is basically the spirit of Section 230, the legal shield that has protected platforms for hosting other people’s words.

The German court said AI Overviews is a different animal. It isn’t pointing you to someone else’s writing through a link. Google’s own AI is generating the sentences, synthesizing the sources, and presenting a single answer in its own voice. So that answer, the court reasoned, isn’t a third party’s speech. It’s Google speaking.

That distinction is everything. Aggregating other people’s links and authoring your own answer carry very different legal weight.

Why This Ruling Lands Now

Step back and you realize the entire mechanic of search has shifted. Old search said: “Here are some relevant pages, you decide.” The burden of judgment sat with the user.

AI search says: “Here’s the answer.” It quietly removes the moment where you’d compare multiple sources. Plenty of people now read the summary and never click a single link. Studies of AI Overviews have shown exactly that — click-throughs to the underlying sites drop sharply once the box appears.

More power pulls more responsibility along with it. If the AI is substituting its judgment for yours, the logic of this ruling says the company that built the AI owns the answer too. Call it the price of the convenience.

Google’s Real Headache

For Google, this isn’t one nuisance lawsuit. It pokes at the foundation of the business model.

If every answer AI Overviews generates can carry legal liability, Google is effectively absorbing billions of potential defamation risks a day. Every time the AI says something wrong about a person, a company, or a product, that’s a possible plaintiff.

But it can’t walk away from AI search either. With OpenAI and ChatGPT chewing into the one market Google has owned for two decades, AI answers are core to survival, not a side experiment. So Google is stuck in the classic bind: it doesn’t want the liability, and it can’t drop the feature.

The twist is that this headache isn’t Google’s alone. ChatGPT, Gemini, Perplexity, and every other service that hands you a generated answer is standing in front of the same question.

What’s Left for the Rest of Us

This is one country’s lower-court decision. It may be appealed. Different jurisdictions may land in different places. But the direction looks clear enough. In an era where AI generates its own words, somebody has to answer for those words.

Maybe we’ve been trusting these answers a little too easily. Maybe we’ve come a long way down this road without ever settling who’s accountable when the answer is wrong. The German court’s value is that it named the empty chair. So the real question is the one worth sitting with: how much do you actually believe the answer your AI hands you?