What happened

When you search on Google, the top of the results often shows an answer written by AI, called an AI Overview; Google also has an AI Mode tab, where AI writes an answer to you directly. This year in Germany, these two kinds of AI answer pinned other people’s wrongdoing on two parties who had nothing to do with it: once on two publishers, once on a doctor. They went to court, and two regional courts ordered Google to stop its AI writing those things, in effect holding Google responsible for what its AI writes.

Both were preliminary injunctions. A preliminary injunction is a temporary order a court issues when a matter is urgent and can’t wait for a full case to run its course, telling the other side to stop in the meantime; it is temporary, not a final ruling.

The first case was in Munich. Two publishers said that when people searched for their name together with the word “scam”, an AI Overview described them as disreputable and getting customers signed up to paid subscriptions without realising it; the publishers said these were other companies’ doings. On 28 May, the regional court in Munich ordered that Google’s AI Overviews must stop saying these things about them. The court’s reasoning was that an AI Overview is new content the AI produces itself after reading several other people’s websites and linking what they say, so it counts as Google’s own statement. Google announced an appeal in June, the parties then settled and the case ended there, so no higher court has ever confirmed that view.

The second case was in Aachen. According to The Decoder and other outlets, when people searched for a local doctor, AI Mode said he had left his job after being accused of sexual coercion of his secretary, and that he had been given a suspended sentence; in fact those events involved a different former colleague at the same hospital. The doctor asked Google to deal with it, and Google replied that it would take no action; even after a letter from his lawyers, the passages were still live. On 24 September, Aachen Regional Court ordered Google to stop its AI writing those things about the doctor, with fines of up to €250,000 for each breach. The case isn’t settled yet: Google can lodge an objection and ask the court to look at it again, so the outcome could still change. As of 30 October, we had seen no report that Google has lodged one.

German courts don’t all agree, though. According to The Decoder and others, a court in Berlin found in a different case that AI Overviews simply gather information from other websites and don’t count as Google speaking for itself. These are all decisions by German courts under German law; they don’t take effect in Taiwan, and there’s no way to tell from them how a Taiwanese court would see a similar case.

Why it matters to you

Say you run a brunch café in Tamsui, New Taipei, relying on local office workers during the week and plenty of day-trippers at weekends. Another brunch place in the same area has a name just one character different from yours, and it recently made the local news after customers reported being sick after eating there. On top of that, you moved from Tamsui Old Street to a spot by the metro station two years ago, some food blogs still list your old address and phone number, and your name on Google Maps is written differently from the one on your website. With old and new information about you mixed together online, and a name one character away from theirs, AI has a hard time telling which pages are about you and which are about them. Suppose a day-tripper asks AI before heading out: “Which brunch place in Tamsui can I trust to be clean?” The AI reads those news reports and writes that your café has had a food-poisoning case. The customer goes somewhere else, and you never find out it happened.

It works the same way in other trades: if you’re a plumber or electrician, complaints about a similarly named business; in a clinic or an office, something a former colleague did; either can end up credited to you by AI. Of the two German cases, one mixed up publishers with other companies and the other mixed up two people from the same hospital, and both ended up in court, where the courts treated them as a real problem. The Munich court’s documents also record a telling detail: later, the same search produced a different AI Overview, which specifically warned readers not to confuse them with another, dubious name. We suspect the original mistake may have come from the AI failing to tell the two apart.

So why can’t AI tell them apart? Our reading is this: when AI writes an answer, it pieces together what it has read on several web pages; the less information there is about you online, the older it is and the more it contradicts itself, the harder it is for the AI to judge which sentences are about you, and the more likely it is to fill the gaps with someone else’s story.

What to do now

Yes: start by finding out how AI describes you right now. In both German cases, the people affected first had to discover they’d been misdescribed, then ask Google to act, have lawyers write and go to court, and it took weeks to months. For a small business, that road is too long, and a court case can only deal with the lines that are already wrong and that you happen to have seen.

When you look, the most important thing is to find out who AI might mix you up with: a business in your trade whose name differs from yours by one character, a shop that opened at your old address after you left, or someone who once worked at the same place as you. Owners often don’t think of this themselves; it means working back from the AI’s answer, opening the source pages listed beside or below the answer one by one to see which of them put the two of you together.

Once you’ve found them, sort things out at the source: fix what you can, and where someone else wrote it and you can’t change it, at least make your own side clear. Make your name, address, phone number and what you do read the same on your website, Google Maps, and food delivery and table booking platforms, and update pages that still show an old address or old name wherever you can. On the reading above, AI may still get things wrong; but with more to tell you apart by, it has less room to fill gaps with someone else’s story, and with your own details the same everywhere, if AI gets something wrong again, the error can only have come from outside, which makes it easier to trace which page led it astray.

You can hand all of this to our managed GEO service. GEO stands for generative engine optimisation, and the engines here are AI tools that write answers on their own, like ChatGPT and Google’s AI Overviews; with managed GEO, our team fixes your site’s content, brings what’s said about you elsewhere into line, and regularly checks how each AI engine describes you; you just review the results.

You can ask AI how it describes you yourself; the catch is that someone else’s story often sits in the same paragraph as the things it gets right, stated just as confidently, and you only pick it out by checking line by line. If you’d like someone to do that with you, test what AI says about you in a demo: a session with one of our consultants where they show you live. The consultant asks ChatGPT about your business name together with words customers worry about, such as “hygiene” or “food poisoning”, then checks the answer against your own information to see which lines are really about someone else. In the same session, the consultant will also show you your website’s GEO audit score. The GEO audit is geoweb.tw’s free website audit, which checks how well AI can understand your site and scores it across 12 areas.

Further reading