The AI answer to your business is now the platform’s own speech. A German court has now said this, and it changes who is liable if the answer is wrong. The lawsuit itself is the smaller story. The bigger issue is what an answer machine does when it can be held accountable for what it says.
The Munich court ruled that the AI overview was Google’s own content
On May 28, 2026, the Munich Regional Court issued an interim injunction (ref. 26 O 869/26) prohibiting Google from repeating false information about two local publishers in its AI overview. The overview had linked them to scams and subscription traps and made connections that did not appear in any of the sources cited.
The court treated the AI overview as Google’s own content and not as a list of search results. In his words, the overview provides “independent, new and substantive statements” through the evaluation and combination of sources, so that the liability protection that applies to an ordinary results page does not apply. It rejected Google’s argument that users should check the answer themselves. When the machine writes the sentence, the owner of the machine stands behind it.
Search engines have always shown false pages, and the law has long protected them from this. The court treated the AI overview as different. It fabricated a false claim by combining fragments from multiple sources into a sentence that contained none of them, and this fabrication the court characterized as authorship. It’s the same recombination that makes AI responses useful: the engine takes your page, rewrites it into something new, and then presents that as a response. A court has now examined the outcome of that trial and described it as “authorized speech,” which carries liability.
The scope here is narrow. This is a regional court, an injunction decided under European liability doctrine, and a U.S. court operating under different speech and intermediary rules could end up elsewhere. In the US, the instinct goes the other way and treats the platform as an immune mediator. This instinct was developed for an era of links and lists, before a machine began writing the sentence itself. It gives direction rather than sets it. This direction sits alongside a realization from a week earlier that being named by an AI does not mean it is believed. Together the two make the form clear. The way an AI response represents your business is both a trust issue and a responsibility issue.
Liability makes the response machine cautious
A response machine that can be held accountable for what it says about a company has every incentive to hedge, tone down, or omit a brand it cannot verify. That is the side effect of the verdict, and it is more important than any individual case. If the answer is the platform’s own speech, the rational response is not to suddenly be accurate. It means being careful.
The companies it can stand behind, that have a consistent, unique, machine-readable identity against which it can base its claims, become the safest names. The blurs become a risk that should even be mentioned.
I don’t know if it’s that clean, and no platform has announced anything like that. But the incentive only points in one direction. Liability makes a system cautious, and a cautious system brings out what it can defend. You can already see the early form of it. Ask an AI about a small or struggling company and watch how often it hedges, relies on an official source, or refuses to characterize the company at all. Liability hardens this reflex from a politeness to a rule. This turns machine-readable identity from a citation tactic into something more at stake. The question is no longer “How do I get the AI to quote me correctly” but is “Am I a company where the AI is confident enough to even mention the name?”
An ambiguous deal is a risk that needs to be mentioned
Most companies give a machine at least one reason to doubt it. Your name resolves into two or three different legal entities on your homepage, in your profiles, and in your old press coverage, and nothing reveals the model, which is canonical. Your founder’s title says one thing on your about page and something else in an interview that the model still trusts. Your product does something specific, but the only place that is clearly stated is within an image or PDF file, which the parser skips. Your category is obvious to a human reading the page and ambiguous to a machine reading the markup, because the page never says in words a parser can understand what the thing actually is.
None of this is a content problem because you’ve learned to think about content over the last decade. It’s an identity problem. The model refuses to make a claim that she cannot properly source, just as a careful editor crafts a sentence that the reporter cannot withstand. This is why accumulating more content as an AI visibility strategy consistently fails. Volume does not resolve ambiguity. A company with ten thousand words and three contradictory descriptions of itself is harder to verify than a company whose homepage reflects in every way the same true statement as a machine reads it. The first looks busy to a person and unreliable to a parser. The second one looks simple to a human and quotable to a machine.
Check what the AI says about you and then correct the facts
You don’t need a lawyer for this. You must be the company that the response machine is confident about.
First, read what the AI is already saying about you. Run your brand, your products, and your category through the search engines your customers actually use and read the responses like a stranger would. Check the specific things a liability-aware engine checks: Does it state your category correctly, assign the right products, name the right people, and avoid associations that aren’t your own? Do it across multiple engines because they will not agree and the spread between them is your test. Most companies have never done this.
Then correct the facts on which the machine is based. Define the entity clearly. Add organizational markup that states who you are, what you do, and how to verify it. Make sure your identity is consistent across all property models read so the engine never has to choose between two versions of you. This is the identity layer of machine-first architecture, the part of the work that makes a business readable by a machine before you ever have to like it. With this ruling, the cost of making a mistake has increased. Not much because it’s still regional, but it’s not nothing.
Then make it a habit and not a one-time exam. Your facts shift, the network around you changes and the models retrain. The companies that remain auditable are the ones that check what the answer says about them on a schedule, just as they would check their own analysis.
The lawsuits will be rare and bound to their jurisdiction. The consequence that matters is slow and structural. If the answer involves risk, the engine becomes cautious, and a careful engine brings out the deals it can stand behind. Do one of these.
Additional resources:
This post was originally published on No Hacks.
Featured image: Viktoriia_M/Shutterstock
Follow us on Facebook | Twitter | YouTube
WPAP (907)
