German court: Google is liable for what its AI invents in search

On May 28, 2026 a court in Munich handed down a ruling that search lawyers will be quoting for years. If a Google AI Overview lies about your business, Google is the one on the hook. Not the source site, not some abstract “algorithm”, but the company itself (the-decoder, Reuters).

It reads like a dry legal headline. It’s actually the first time a court said it plainly: an AI answer in search is not “search results”, it’s the search engine’s own words. And it answers for those words as an author, not as a library that just handed you someone else’s book.

What actually happened

Two publishers from Munich went to court because a Google AI Overview tied them to fraud, subscription traps and “dubious business practices”. The neural net claimed they were “known for” this. One problem: those links appeared in none of the sources the overview itself cited. The AI simply invented them and glued real names to shell companies.

The Munich Regional Court (case 26 O 869/26) banned Google from showing the claims and put 80% of the legal costs on the company. Google disagrees and has filed an appeal (Reuters). Its position: AI Overviews “reflect information that exists on the web” and run at 91% accuracy.

That’s where the whole plot hides. 91% sounds solid until it’s about you. If the AI is wrong one time in ten, and the error is “known for fraud” next to your name, the percentage stops being comforting.

Why this is a turning point, not a routine lawsuit

Search engines used to shut every claim down with one line: “We don’t assert anything, we show third-party results.” The entire defense rested on it - the operator answered not as an author but, at most, as an intermediary obliged to remove a link after a complaint. Germany’s Federal Court of Justice held that line for years.

The Munich court broke it. The logic is simple: an AI Overview doesn’t display or link to someone else’s text. It rewrites and judges the results in its own words and its own structure, producing independent, new statements that aren’t in the search results at all (DW). Since only Google writes and controls the algorithms, Google is the author of those words.

The court rejected the “users can verify it themselves” defense. Rightly so: the whole point of an AI Overview is that the person never clicks away. The answer is right there, ready-made. One in a hundred will go and check.

The line between “search engine” and “publisher” moved that day. Where search became a neural net’s answer, the search engine became the author of that answer.

What this has to do with you and your business

This is Germany, not your country, and the ruling is still under appeal. It has no direct legal effect on your market. But the thing to watch isn’t the jurisdiction - it’s the mechanism.

The mechanism is the same everywhere. A neural net pulls fragments from the index and retells them in its own words. Google calls it AI Overviews, others have their own AI answers, and both already sit on top of part of the results page. Either one can glue your brand to something it has nothing to do with - simply because similar words sat nearby in the index.

The old marketer’s pain was “losing traffic”: the AI answered, the person never reached the site. The new pain is sharper. The AI can state an invented fact about you, and the customer reads it as truth without opening a single link. Reputation now forms in a block you don’t control and can’t always even see.

I track what AI pulls from my own projects . On a cleaning project with history and systemic SEO, Copilot cites the pages 40 times in six weeks. On a fresh personal site - zero. The conclusion sobered me up: whoever the model can’t see in proper sources, it makes up. It doesn’t leave the gap empty - it fills it with a guess.

What to do right now

Don’t panic, and don’t wait until a court in your own country gets there. Do four things.

1. Check what AI says about you. Ask ChatGPT, Perplexity, Copilot and your local search AI about your company, name and services. Write down where it lied or invented something. Five minutes, and the picture is often ugly.

2. Control your footprint in the index. AI cites what sits well in search: index depth, factual pages, a link profile. With no clear data about you, the model builds an image from scraps. Filling that gap with facts is your job, not its.

3. Give the machine a factual minimum. Who you are, what you do, where you work, contacts - in machine-readable form: Schema markup (Organization), a clear “About” page, an llms.txt for the models. So the AI has something to take besides guesses.

4. Monitor regularly, not once. AI answers shift with the index. I track AI visibility monthly through the Bing AI Performance report and check what gets cited and from where. A one-off check shows a snapshot; a regular one shows the trend.

That’s what GEO is - optimizing for the answers of neural nets. Not magic, not a separate trendy channel, but hygiene: watch what the machine says about you, and feed it the right sources.

Bottom line

The Munich ruling is formally about Google and about Germany. But the signal is wider, and it touches anyone with a name online. The era of “the search engine isn’t responsible for its results” ended where the results became a neural net’s answer.

For business the takeaway is short. Stop treating AI answers as someone else’s responsibility. That’s what your customers read about you before they ever reach your site. And if it’s empty or wrong there, the model will write the rest itself.

Want to know what AI search says about your business and strengthen your footprint in the sources - message me on Telegram @dimik90 or order an audit. I’ll look at what the machine says about you and why, and what to do about it.

How I run SEO and AI visibility on a live project is in the cleaning case with SEO from zero in 50 days. And why marketing drowns in operations instead of results is a separate breakdown from a real project.