AI Search Overviews Ruled Editorial Content in Germany — What Businesses Must Do

Reading time: ~5 minutes.

TL;DR: On July 16, 2026 Germany’s media regulator, ZAK, announced it will treat AI search overviews and chatbot answers as the platforms’ own editorial content, which, ZAK says, places those outputs under the Interstate Media Treaty (Section 109) rather than under the Digital Services Act’s intermediary protections. The decision signals new transparency, registration and accountability expectations for AI providers and changes the commercial and legal calculus for publishers and enterprises using generative systems.

A legal classification that changes the game

ZAK (applied the Interstate Media Treaty to AI‑powered search overviews) announced on July 16, 2026 that it has applied the Interstate Media Treaty to AI‑powered search overviews and chatbots, finding violations of Section 109. Dr. Thorsten Schmiege, ZAK’s chair, put the regulator’s position plainly:

“AI search engines and chatbots are content providers, and we are now consistently applying German media law to them.”

ZAK says these rulings are immediately enforceable and that Google and Perplexity have one month to appeal. ZAK also concluded that the EU Digital Services Act’s intermediary liability protections do not apply where an operator’s system produces outputs treated as the provider’s own editorial content. The regulator frames that legal point as central to its orders. How higher courts or EU institutions will ultimately interpret these boundaries remains an open legal battleground.

Why the regulator draws that line

ZAK’s reasoning rests on three linked ideas.

Editorial character: Regulators and a Munich court have found that AI summaries can produce “independent, new, and substantive statements” rather than mere pointers to third‑party material. When a model synthesizes reporting into a single narrative or asserts facts in its own prose, authorities treat that as the platform’s editorial expression.
Transparency and discovery effects: ZAK flagged that prominent AI summaries can push traditional journalistic links down the page and materially reduce referral traffic. Dr. Schmiege warned that control over selection and placement of links must be transparent to preserve media diversity.
Traceability and liability: The argument follows that if the output is the operator’s message, it cannot claim the DSA’s host or redistributor shield in the same way. That exposes platforms to publisher‑style obligations and potential liability for false or harmful statements.

Traffic reality: users rarely click the source link

The commercial risk behind ZAK’s concern is measurable. Pew Research Center reported on July 22, 2025 that, in a metered browsing sample of U.S. desktop and mobile traffic collected in March 2025, users clicked a source link inside Google’s AI overview in only about 1% of visits where an overview appeared. Pew also found overall clickthrough rates were lower when a summary was shown (roughly 8% with a summary versus 15% without, in that sample).

That pattern, readable answers reducing outbound clicks, is the empirical engine driving publishers’ alarm: less referral traffic means smaller audiences and revenue, which in turn feeds the regulatory urgency to preserve plurality and discoverability.

The Munich decision and supporting analysis

A regional court in Munich (case no. 26 O 869/26, June 2026) reached similar conclusions in litigation against Google, describing AI responses as producing “independent, new, and substantive statements” and treating those outputs as the operator’s content. The court issued provisional relief in that dispute; Google says it will appeal.

Empirical critiques bolster the legal reasoning. Reporting that cites an Oumi analysis (as covered by The Decoder and the New York Times) found that Google’s Gemini 3 Overviews achieved high measured accuracy in some tests (reported as ~91%), yet a large share of those correct answers could not be directly corroborated by the links shown, raising traceability questions when platforms are asked to justify assertions at scale.

Google, Perplexity and the product response

ZAK’s announcement named both Google and Perplexity. According to ZAK, Perplexity lacked required transparency disclosures and did not have a designated representative in Germany, a practical compliance gap if the service is to be treated as a media provider. (ZAK is the source for those specific findings.)

Google has deployed product features such as “Preferred Sources” to let users or partners prioritize particular outlets, and the company has argued in public statements that AI Overviews reflect existing web information and aid discovery. Regulators and the Munich court say product knobs alone won’t resolve the underlying questions about editorial responsibility, transparency, and the downstream impact on publishers.

What this can mean for businesses and publishers

If national regulators succeed in treating AI outputs as the operator’s own editorial content, practical consequences could include:

Registration, transparency and non‑discrimination obligations under media law (ZAK cites Section 109 of the Interstate Media Treaty, which governs registration and transparency duties for media providers).
Requirements to appoint local representatives and maintain traceability and recordkeeping for editorial decisions and source selection.
Increased exposure to legal claims for false or defamatory assertions, and to administrative enforcement measures, the exact remedies depend on ZAK’s orders and any later court rulings.
Material commercial pressure on publishers, who may need new licensing, structured data or negotiated agreements with AI providers to recoup lost referral value.

There is pushback. Platforms argue that summaries improve user discovery and that the Digital Services Act was designed to balance platform functionality with liability protections. Those arguments set up a likely, prolonged clash between national media‑law enforcement and EU‑level interpretation of intermediary rules.

How this could play out across the EU

The DSA provides a Europe‑wide regulatory framework for online intermediaries, but it does not fully resolve whether operator‑generated syntheses sit inside or outside intermediary safe harbors. That question will be tested in national courts and possibly referred to the Court of Justice of the EU. Expect litigation and rulemaking to shape whether ZAK’s approach becomes a broader precedent or a national outlier.

Practical steps leaders should prioritize

Treat this as a regulatory and commercial risk that requires immediate triage and a strategic plan. Below are pragmatic, prioritized actions with suggested deliverables and approximate resource intensity.

Immediate (next 30-90 days)

Run a “top prompt” factual‑risk audit. Deliverable: spreadsheet of the top 100 prompts producing assertive claims and a risk score. Resource intensity: low, medium.
Label AI outputs in customer channels. Deliverable: a clear UI flag that an answer is AI‑generated and a short “how this answer was produced” tooltip. Resource intensity: low.
Designate legal representation in key jurisdictions. Deliverable: appointment letters and contact info for local reps where products are active. Resource intensity: low.

Near term (3-9 months)

Implement provenance and citation logging. Deliverable: logs that map answers to model prompts, training data slices or source links for audit. Resource intensity: medium, high.
Negotiate publisher arrangements. Deliverable: pilot licensing or attribution deals with top content partners and metadata-sharing agreements. Resource intensity: medium.
Harden human‑in‑the‑loop for high‑risk domains. Deliverable: workflow that routes legal/medical/financial queries to expert review before publication. Resource intensity: medium.

Strategic (9-18 months)

Refactor product affordances to reduce legal exposure. Deliverable: product roadmap that limits assertive AI claims in consumer‑facing answers or requires source verification for certain classes of statements. Resource intensity: high.
Policy engagement and litigation readiness. Deliverable: legal playbook, regulatory monitoring dashboard, and budget for potential appeals or defense. Resource intensity: medium, high.
Measure and monetize discovery impact. Deliverable: analytics linking AI answer placement to referral revenue and a commercial strategy (licensing, revenue‑share, or API access tiers). Resource intensity: medium.

Three short, practical questions, and blunt answers

Does this change legal exposure for platforms and integrators?

ZAK says yes: treating AI outputs as the operator’s editorial content removes the intermediary shield the regulator associates with redistributors. How binding that view becomes across the EU depends on appeals, national court rulings, and possible CJEU guidance.
What should publishers do to protect revenue?

Push for transparent citation and stronger negotiating positions: test licensing pilots, insist on structured metadata that helps trace original reporting, and explore pay‑for‑index or referral‑share models. Short term, track referral deltas where AI summaries appear and use that data in commercial talks.
What must enterprise product teams do now?

Prioritize auditability and provenance: tag outputs, log prompts and sources, and enforce human review for high‑risk content. Also appoint local legal contacts and build a regulator‑facing compliance pack so you can respond quickly if an authority asks for records.

Final thought

Germany’s move reframes a design choice, showing that putting an answer on the page is now a regulatory and commercial decision. For executives, the practical question is no longer just “can we build helpful assistants?” but “how will we govern what they say, prove where answers come from, and compensate the ecosystem that created the raw material?” Regulators, publishers and platforms are now bargaining over that answer in courtrooms and negotiation rooms alike. Act accordingly: treat your models as if they already carry a byline.