Germany ruled AI search answers are media, not a neutral pipe

The question underneath every AI search fight is deceptively simple: when an AI engine writes an answer, is that the engine's speech, or is it just plumbing? Germany has now answered in two directions at once - and the disagreement is more instructive than either ruling alone.

For a decade, platforms have relied on a comfortable legal position: we do not write the content, we only organise it. That framing is what keeps liability at arm's length. AI answers strain it badly, because the engine really is producing the sentence.

Germany spent seven weeks in 2026 pulling at exactly that thread.

Three decisions, not one

They are separate events with different actors, and they are easy to blur together. It is worth keeping them apart.

28 May 2026, Regional Court of Munich I. In case 26 O 869/26, the court ordered Google to stop disseminating false claims about two Munich-based publishers that AI Overviews had linked to fraud and subscription traps, with penalties up to 250,000 euros per violation. The court's reasoning was the striking part: it treated the AI-generated statements as Google's own statements rather than third-party content it merely surfaced. Note that this came through injunction proceedings, so it is best described as a court order rather than a final merits judgment.

1 June 2026, Regional Court of Berlin. Four days later, a different German court went the other way, dismissing a trademark claim on the reasoning that Google does not present AI answers as its own statements, treating them instead as a new search format. Same country, opposite premise.

14 July 2026, ZAK. Then the regulator arrived. ZAK, the joint commission of Germany's state media authorities, acting on investigations by the Hamburg/Schleswig-Holstein and Berlin-Brandenburg authorities, ruled that Google AI Overviews and Perplexity operate as content publishers, not neutral conduits, applying the State Media Treaty to AI search output. It is the first time a European media regulator has applied broadcasting-style media law to AI-generated search results.

"The interesting fact is not that Germany decided. It is that Germany decided twice, in opposite directions, within four days."

Curious how AI engines describe your brand right now? Get a free visibility audit and see where you stand across ChatGPT, Gemini and Perplexity.

Why the split is the story

Coverage of the Munich decision tended to declare the matter settled: AI answers are Google's own words, therefore Google is liable. That overstates it. Google confirmed on 12 June 2026 that it will appeal, arguing the case focuses on specific and narrow errors rather than the foundational way AI Overviews displays web content. The judgment is not legally final. And the Berlin ruling shows German courts have not converged on a single view.

What is genuinely established is narrower and still significant: the neutral-conduit defence is no longer automatic in Germany. A regulator has said it does not hold under media law, and at least one court has said it does not hold under personality rights. Whether that survives appeal is open.

What it changes for everyone else

The second-order effects matter more than the legal specifics, especially for brands outside Germany.

• Accuracy becomes a liability question. If an engine can be held to publisher standards for what it asserts, hallucinated claims about a company stop being an embarrassing bug and start being a legal exposure.
• Grounding gets more valuable. The cheapest defence against saying something false is to anchor answers in sources that can be pointed at. That pushes engines toward well-structured, clearly-attributed, internally-consistent sources.
• Being wrongly described is now actionable somewhere. The Munich publishers did not ask for better SEO. They asked a court to stop an AI system misrepresenting them, and got an order. That is a new lever for brands facing AI misinformation about themselves.
• Fragmentation is coming. Germany applies media law, the UK mandates opt-outs and attribution, the EU AI Act adds disclosure duties from 2 August. Engines will increasingly behave differently by jurisdiction.

The takeaway

Germany has not settled whether AI answers are media. It has demonstrated that the question is live, contested, and being decided by bodies that do not agree with each other. For brands, the practical response is not to wait for the law to resolve. It is to make sure that when an engine describes you, it has accurate, consistent, easily-verified material to draw on - because the direction of travel, in every jurisdiction, is toward answers that someone can be held responsible for.

What this does not mean

Because the Munich order was the loudest of the three events, it is easy to walk away with conclusions the record does not support.

It does not mean AI search is now illegal in Germany, or that engines must stop generating answers. The regulator's finding is about how these products are classified, not whether they can exist. Media law brings obligations around transparency and accuracy; it does not amount to a ban.

It also does not mean a brand can force an engine to say something flattering. The lever the Munich publishers used was narrow: they showed a specific, damaging falsehood and asked for it to stop. That is a remedy against being wrongly described, not a right to be described well. A company that simply dislikes how it is summarised has no equivalent claim, in Germany or anywhere else.

And it does not mean the law has settled on your side. The Berlin court reached the opposite premise, the Munich order is under appeal, and a regulator's classification is not a binding court judgment. Treating any one as the final word is exactly the mistake the seven-week split warns against.

What to actually do about it

The useful moves while the law is unsettled are the same ones you would make regardless of how any court rules. They cost little, they compound, and none of them depend on Germany.

• Find out what the engines currently say. Ask each major engine plain questions about your company - what it does, who it serves, whether it is legitimate - and record the answers verbatim. You cannot correct a description you have never read.
• Fix the source, not the symptom. When an answer is wrong, trace it back to what it is grounded in. A stale directory listing, an old press mention, or a thin "about" page is usually doing more damage than any single hallucination.
• Make the true version easy to reach. Put the facts you want repeated - founding, location, category, what you are not - in clear, consistent language across the pages an engine is most likely to trust. Contradictions between your own sources are an open invitation to guess.
• Keep a dated record. If you ever need to act on a damaging AI claim, the first thing anyone will ask is what was said and when. A running log of engine outputs turns a vague complaint into evidence.

None of this requires a legal department or a settled jurisdiction. It is ordinary hygiene that becomes more valuable the moment accuracy carries consequences.

How to know if it is working

Publisher-grade accountability rewards being consistently, checkably right about yourself. That is measurable, and the measurement is simpler than most brands expect.

Pick a short set of questions a real customer or investor might type, and ask them across several engines on a fixed schedule. Watch two things over time: whether the engine names you where it should, and whether what it says lines up with what is true. A brand that is well grounded gets named more often and misdescribed less. A brand that is not drifts in and out of answers and picks up errors that nobody planted.

"You do not manage what an engine says by arguing with it. You manage it by controlling what it has to read."

The point of tracking is not vanity. It is to catch a wrong answer while it is still one engine's mistake, before it hardens into the version everyone repeats. As the direction of travel moves toward answers someone can be held responsible for, the brands that come out ahead already know what is being said about them, and have made the accurate version the easiest one to cite.