The Ruling
On 14 July 2026, Germany's Commission for Licensing and Supervision (ZAK) — the body coordinating the country's 14 state media authorities — issued the first rulings anywhere classifying AI search summaries as the provider's own editorial content rather than intermediated third-party material. The proceedings, led by the Hamburg/Schleswig-Holstein and Berlin-Brandenburg media authorities, targeted Google's AI Overviews and Perplexity's AI search assistant. ZAK found that Google's AI-generated summaries, placed prominently above organic search results, impermissibly push down and discriminate against journalistic links — a violation of the Medienstaatsvertrag's non-discrimination obligations for media intermediaries. Perplexity was separately flagged for lacking a required German legal representative and transparency disclosures (die-medienanstalten.de). "AI search engines and chatbots are content providers, and we are now consistently applying German media law to them," ZAK chairman Dr. Thorsten Schmiege said (The Decoder). Both rulings are immediately enforceable, with a one-month window to appeal; Google has said it will do so.
The legal move underneath the headline matters more than the fine print. ZAK's theory is that the EU Digital Services Act's hosting-liability shield — which protects platforms from liability for third-party content they merely store or display — does not cover AI-generated answers, because an AI Overview is not third-party material passed through a pipe. It is text the model wrote. That is a coherent reading of DSA Article 6, which conditions the shield on the provider not exercising "authority or control" over the content in question (European Commission).
Steelmanning ZAK
The regulator has a real point, and it deserves to be stated plainly before it's argued against. Google did not build AI Overviews as a passive index of the web; it built a system that reads sources, synthesizes an answer in Google's own voice, and displays that answer above the very links it drew from — while giving publishers no meaningful opt-out that doesn't also cost them ordinary search visibility. If a summary is wrong, defamatory, or simply crowds out the reporting it was built on, "we're just hosting third-party content" is a strained defense for text the system authored. Traditional publishers who write similarly synthesized copy don't get a liability shield; treating an LLM's synthesis as functionally different because a model produced it, rather than a person, is a distinction regulators are entitled to test. Perplexity's alleged failure to designate a German representative is not a close call at all — that is a bog-standard DSA compliance obligation, and "we're an AI company" is not an exemption from it.
Where the Ruling Overreaches
The discrimination theory is where this gets shakier. ZAK's complaint is functionally that Google organizes its own results page in a way that favors Google's own feature — which is a competition and self-preferencing question, the kind of thing the EU already built an entire regime (the Digital Markets Act) to police through ex-ante conduct rules and ongoing Commission enforcement against designated gatekeepers. Routing that same complaint through 14 state media regulators applying a national broadcasting-descended statute, the Medienstaatsvertrag, risks a patchwork where Germany's answer to "is AI Overviews anticompetitive self-preferencing" diverges from Brussels' answer, decided under different standards, by different institutions, with different appeal paths. That's not hypothetical friction — it's exactly the fragmentation the DSA's country-of-origin principle and one-stop-shop enforcement model were designed to prevent for platforms operating across 27 member states.
There's also a version-control problem baked into treating a live AI answer as fixed "content" subject to accuracy and plurality obligations. A newspaper article is a discrete artifact you can audit once. A generative answer is produced fresh, differently, for every query and every user, at a scale no publisher-style review process was built to handle. Holding the system design — does it push links down, does it disclose sourcing, does it name a legal representative — to a compliance standard is tractable. Holding each individual output to editorial-content liability norms invites either massive over-caution (models refuse to summarize anything contested) or liability exposure so broad it becomes a de facto tax on offering an AI answer feature in Germany at all, borne disproportionately by smaller entrants like Perplexity who can't absorb it the way Google can.
What Comes Next
Google's appeal will test whether a national media regulator can carve an AI-specific exception into the DSA's liability architecture without a preliminary reference to the Court of Justice of the EU — DSA liability questions are exactly the kind of interpretive dispute the CJEU exists to settle uniformly across the bloc. If ZAK's reasoning survives appeal, expect France's Arcom, Italy's AGCOM, and other national regulators with parallel media-pluralism mandates to test the same theory against the same handful of AI answer engines, each under its own domestic statute. That is a plausible and, for publishers who've watched referral traffic erode, an understandable outcome. It is also the definition of regulatory fragmentation the DSA was supposed to end — replaced, in this reading, with 27 national media laws applied AI-service by AI-service. The fix Brussels should actually pursue is a DSA-level answer on what counts as "the provider's own content" for generative outputs, decided once, not fought out state authority by state authority while Google's engineers wait to see which country's answer sticks.