German Court Holds Google Responsible for What Its AI Says About You
AI feedback about your business is the talk of the speaker itself now. The German court has said that, and it changes who is responsible if the answer is wrong. The case itself is a small matter. Bigger is what the response engine does when it can be held accountable for what it says.
Munich Court Ruling That Google’s Own Content AI Overview
The District Court of Munich issued a temporary injunction on May 28, 2026 (case 26 O 869/26) preventing Google from repeating the false statements its AI overview made about two local publishers. The overview was tied to scams and subscription loops, drawing connections that don’t come from one of the identified sources.
The court treated AI Overview as Google’s own content rather than a list of search results. In its own words, the overview produces “independent, fresh, and robust statements” by measuring and combining sources, so that the credit protection that covers the standard results page is ineffective. It rejected Google’s argument that users should check the answer themselves. When a machine writes a sentence, the owner of the machine stands behind it.
Search engines have always displayed bad pages, and the law has protected them for a long time. The court treated the AI Overview as unique in nature. He created a false claim, combining fragments from several sources into a sentence that did not belong to them, and that production was what the court called ownership. It’s the same combination that makes AI responses useful: The engine takes your page and rewrites it into something new, then presents that as a response. The court has now looked at the outcome of that program and called it authorized, criminally charged speech.
The range here is narrow. This is a single regional court, a temporary order, decided under the European debt doctrine, and a US court acting on separate speech and arbitration rules can come elsewhere. In the US, the instinct runs the other way, in treating the platform as an immune mediator. That thinking was built in the era of links and lists, before the machine started writing the sentence itself. It points to a direction rather than solving one. That guidance sits alongside last week’s finding, that being invented by AI doesn’t mean believing in it. Together, these two make the situation clear. How AI feedback represents your business is an issue of trust and an issue of accountability at the same time.
Credit Makes the Answer Engine Vigilant
An answer engine that might be responsible for what it says about a business has every incentive to hedge, soften, or exclude a brand it can’t verify. That is the result of the second phase of the decision, and it is more important than any single case. If the answer is the speaker’s own speech, the logical answer need not be suddenly accurate. It is awareness.
Businesses that can stand behind them, those that have a consistent, unambiguous, machine-readable identity that can back up their claims, are the ones that are safe to call. The unintelligible become dangerous to say at all.
I don’t know if it plays this clean, and no forum has announced anything like it. But encouragement only points one way. Liability makes the system vigilant, and a vigilant system exposes what it cannot protect. You can already see its original shape. Ask the AI about a small business or competitor and watch how often it hedges, backs away from a legitimate source, or refuses to reveal the company at all. Obligation hardens that thinking from respect to law. That turns machine-readable identity from a citation trick into something closer to table stakes. The question stops being “how do I get the AI to quote me correctly” and becomes “am I a business that the AI is confident enough to name at all.”
Unclear Business is Risky You Can’t Say
Most businesses give the machine at least one reason to be skeptical. Your name resolves to two or three different legal entities on your home page, your profiles, and your past press releases, and nothing tells the model which canonical book it is. Your founder’s title says one thing on your About page and another in an interview the model still trusts. Your product does something specific, but the only place it’s clearly mentioned is within the image or PDF that the reviewer skips. Your category is obvious to a page reader and incomprehensible to a mark-reading machine, because the page never says, in the words the parser might suggest, what the object really is.
None of this is a content problem in the way the last decade has trained you to think about content. It’s an identity problem. The model refuses to make the claim that it cannot appear cleanly, the way a careful editor strikes a sentence a journalist cannot stand up. This is why stacking more content ends up failing as an AI visibility strategy. Volume does not resolve ambiguity. A business with ten thousand words and three conflicting definitions is harder to verify than a business whose homepage says the same true thing in every machine-readable way. The first one looks busy to the person and dishonest to the analyst. The second seems clear to man and affects the machine.
Check What AI Says About You, Then Correct The Facts
You don’t need a lawyer for this. You need to be a business that the response engine is sure of.
Start by learning what AI already says about you. Run your brand, your products, and your category on the engines your customers use, and read the responses like a stranger would. Check some things that a credit-aware engine will check: does it state your category correctly, describe the right products, name the right people, and avoid organizations that are not yours. Do it in all the engines, because they will not agree, and the spread between them is your research. Most businesses never do this once.
Then correct the facts the basis of the machine. Define the business clearly. Add an organizational tag that says who you are, what you do, and how you can be verified. Keep your identity consistent across all read structures, so the engine never has to choose between your two versions. This is the Identity layer of the Machine-First Architecture, the piece of work that makes a business machine-readable before it likes you. The cost of getting it wrong increased with this decision. Not much, because it’s still a district, but it’s nothing.
Then make it a habit, not a one-off audit. Your facts are moving, the web around you is changing, and models are retraining themselves. Businesses that always confirm are the ones that check the feedback on the schedule, the way they would check their analytics.
Cases will be rare and confined to their areas. The main effect is slow and structured. When the response is risky, the engine is aware, and the careful engine looks for businesses that can stand behind. Make your own one of them.
Additional resources:
This post was originally published on No Hacks.
Featured image: Viktoriia_M/Shutterstock
