When Google's search box was a list of ten links, Google could say it was only pointing at other people's pages. Two German courts have now decided that defence does not cover the paragraph an AI writes above those links.
The Decoder reported both cases on October 18. Neither ruling is final, and Germany is one country. But both describe the same thing anyone who tracks AI answers sees every week: the engine is not quoting a source, it is writing something new about you.
Munich: an AI Overview is Google talking
In late May, the Munich Regional Court I heard a case brought by two Munich publishers. Google's AI Overviews had linked them to fraud schemes they had nothing to do with.
The court held that an AI Overview is Google's own statement, so Google is liable for what it says. That is a break from how search has been treated for twenty years, where Google was a neutral intermediary passing along what third parties had published.
Google announced an appeal and then settled instead. So no higher court has ruled on the reasoning, and nobody outside the case knows whether money changed hands.
Aachen: a fabricated scandal, and a refusal to remove it
The second case is the one I would show a founder. A professor of neuroradiology left his post as chief physician at the end of June 2026, after holding it since 2010. On August 28 a colleague told him that Google's AI Mode was explaining his departure with an alleged sexual assault on his secretary.
Searches with words like "suspension", "termination" and "court proceedings" produced several paragraphs of detail, including a legal aftermath that never happened. The accusation and the proceedings belonged to a different former professor. The court could check that against its own files, because the criminal case had been against someone else.
He emailed Google in late August asking for removal. In mid-September Google said it would take no action, and a formal legal demand the same day changed nothing. On September 24 the Aachen Regional Court issued a preliminary injunction against three versions of the text, without hearing Google first, with a penalty of up to €250,000 per violation.
The court's reasoning is worth quoting: Google "provides AI-generated search and is therefore also responsible for the answers generated in this context." It added that Google could have checked the output against its own sources.
Why the error happened is the useful part
Strip out the legal question and look at the failure. The engine had two real people with similar profiles, both former professors, and it merged them. Every fact it used probably existed somewhere on the web. The answer attached them to the wrong name.
That is the same failure we see in smaller form when we run buyer questions through ChatGPT, Claude and Google's AI answers for B2B startups. A company gets confused with another one that has a similar name. A product page from 2023 is read as current. A competitor's feature is credited to the wrong brand. Nobody sues over those, but a buyer who reads them may still cross you off a shortlist.
Where I am not convinced
A lawsuit fixes one answer, after the harm. The doctor needed a colleague to notice, two months of emails and a court to stop three specific texts. AI Mode can generate a fourth version the next time someone asks a slightly different question, and the article itself notes that liability for answers that are generated constantly is hard to even measure.
Settlements also cut against precedent. If Google keeps settling the cases it might lose, there may be no top-court ruling for years, and outside Germany these decisions bind nobody. I would not plan around the courts making AI answers accurate.
The bigger picture
What these rulings settle, at least in two German courtrooms, is the framing: an AI answer about your company is a statement someone made, and it can be wrong in ways a list of links never was. You cannot correct it by editing the answer, because there is no fixed answer to edit.
The practical move is the boring one. Ask the engines the questions your buyers ask, read what they say about you, and when an answer is wrong, find the pages it is built from and fix or outweigh those. Clear, consistent pages about who you are and what you do give the engine less room to merge you with someone else.
ClawWorld runs AI visibility work for B2B startups. Start with a free baseline report: 10 buyer questions, who ChatGPT and Claude name instead of you, and the pages they cite.