Munich: Google's AI Overviews Are Google's Words

A Munich court just drew a line Silicon Valley hoped would stay blurry: when Google's AI Overviews invent false claims about your company, Google said them — not the websites it cited. The ruling reclassifies generative search from neutral indexing to publisher liability, and the "created with AI" label offers no escape hatch.

What happened:

• On May 28, 2026, the Landgericht München I issued a preliminary injunction in case 26 O 869/26 against Google.
• Two Munich publishers — Verlagshaus24 and GeraMond Verlag — sued after AI Overviews falsely linked them to fraud schemes, subscription traps, and shady business practices.
• The claims appeared when users searched the company names alongside "Betrugsmasche" (fraud scheme).
• Google must stop repeating the false statements and bear 80% of court costs. Violations carry fines up to €250,000 per instance.

What the AI did wrong:

• Searches on January 20 and 26, 2026 returned structured AI answers opening with assertions that the publishers were "known for disreputable business practices."
• The court found the AI fabricated connections between the plaintiffs and other sketchy companies — connections that did not appear in any cited source page.
• A third search on February 10 — after Google received a formal cease-and-desist letter — returned substantially similar content.
• Google's own reporting form redirected the complaint; the harmful output persisted anyway.

The legal reclassification:

• Conventional search results display indexed third-party snippets with links. AI Overviews synthesize, structure, and present an independent answer — and users read it as Google's voice.
• The court rejected Google's argument that it was merely a hosting provider under the EU Digital Services Act. The content was generated by Google's own system.
• It also rejected search-engine immunity: AI Overviews are "by no means indispensable for internet use" the way ranked links are, so they do not inherit the reduced liability German courts have granted search since the early 2000s.
• A "created with AI" disclaimer does not transfer verification burden to the reader. Under German law's "Titelseitenleser" standard — casual front-page readers — the overview reads as a complete, self-contained statement.

Why the link defense failed:

• Google argued users could click source links and verify the claims themselves.
• The court rejected that under Germany's "Titelseitenleser" standard — casual readers treat the overview as a complete answer, not a prompt to fact-check.
• Independent data explains why: a Pew Research Center study of 900 U.S. adults in March 2025 found users click a source link inside an AI Overview in just 1% of sessions — and overall click-through to any result drops from 15% to 8% when a summary appears.
• The answer satisfies the query inside Google's interface. Almost nobody audits it.

The precedent chain:

• A Frankfurt court ruled in September 2025 (case 2-06 O 271/25) that AI Overview liability is not per se excluded — but dismissed that specific injunction on the facts.
• Munich went further: it prohibited specific false statements, not just affirmed theoretical liability.
• University of Amsterdam law professor Natali Helberger, who directs the AI, Media & Democracy Lab, called the ruling significant on June 10 — noting four pillars: structural difference from search results, Google's sole control over the technology, the absence of similar results in conventional search, and the non-essential nature of AI Overviews.

The US contrast:

• Section 230 of the Communications Decency Act still shields US platforms from liability for third-party content.
• Germany has no equivalent for AI-generated synthesis. The Munich court treated the output as original speech by the operator.
• Parallel US suits are building on different frameworks — Wolf River Electric sued Google in March 2025 after Gemini fabricated a non-existent attorney general lawsuit; activist Robby Starbuck filed a $15 million defamation suit in October 2025 over false AI-generated claims.
• The EU's draft AI Liability Directive is being negotiated with exactly these questions in mind. Munich just gave negotiators a live precedent.

What changes for operators:

• Publishers and companies in Germany can now seek preliminary injunctions against false AI Overviews with a lower hurdle — attribution runs to Google as direct infringer, not passive intermediary.
• Strategically, the ruling pressures Google to run AI Overviews with more conservative parameters on sensitive topics: individuals, companies, health.
• For every enterprise deploying AI that presents synthesized answers under its own brand — not just Google — the logic travels: if you generate it, you own it.

Convina's view: Silicon Valley spent twenty years arguing that platforms are pipes, not publishers. Generative search broke that fiction in one product cycle — and a Munich civil chamber just wrote the obituary. The question was never whether AI would hallucinate. It was who pays when the hallucination lands on a company's reputation inside an interface billions trust as ground truth. Europe answered first. US lawmakers are still debating whether to regulate model weights. Every organization that ships AI-generated answers to customers, employees, or citizens should read this ruling as an operating mandate: provenance, human review, and kill switches are not nice-to-haves — they are the new cost of calling something yours.