Google is responsible for what its AI search summaries say, a German court has decided, in one of the first serious tests of who answers when a generative system gets it wrong. The Regional Court of Munich granted a preliminary injunction against Google after its AI Overviews wrongly tied two Munich publishers to scams and shady business practices — connections that appeared in none of the sources the summary cited.

The ruling matters because of how the court reached it. Rather than treat the AI Overview as a neutral list of results — the legal status that has long shielded search engines from liability for third-party content — it classified the summary as Google’s own content, written in its own words. For any brand that has watched an AI answer invent something damaging about it, that reclassification is the consequential part: it puts the liability on the platform that generated the claim.

What the Munich court actually ruled

The Regional Court of Munich issued a temporary injunction, case 26 O 869/26, ordering Google to stop its AI Overviews from repeating false claims about the publishing house Verlagshaus24 and a subsidiary. The court called Google a direct infringer, finding that its AI rewrites and restructures search results into new statements of its own, opening with confident assertions — in one instance, that a company was known for dubious business practices — then building a summary, a list of supposed red flags and tips for users. Some of those claims, the court found, were not made in the linked results at all; the system had mixed the publishers up with genuinely fraudulent firms.

Google’s defence — that users can click the cited links and check for themselves — was rejected. The option to fact-check a statement later “does not, as a rule, exempt one from liability” for it, the court held, noting that the feature’s convenience collapses if every link must be verified by hand. That point lands harder against a Pew finding that barely 1% of users click through from an AI Overview to its sources. Google said the summaries are designed to reflect information already on the web, that it invests heavily in their accuracy, and that it is reviewing a decision it stresses is “not yet final”.

What it means for brands and their reputations

For marketers, the ruling is a rare piece of leverage. AI answers, Google’s and others’, can state as fact things no source supports, and until now the route to a correction has been unclear. Treating the summary as the platform’s own speech gives a wronged brand somewhere to direct a complaint and a legal theory to back it.

The practical implication is to watch what AI search says about a brand as closely as its own website. That means querying the brand and its products in AI Overviews and assistants, capturing fabrications and false associations as they appear, and escalating them — the Munich publishers sent a cease-and-desist before going to court, and Google’s failure to respond adequately formed part of the case. Screenshots with timestamps and the exact query matter here, because an AI answer can change between sessions and a fabrication is hard to act on once it has vanished. The caution is real: this is a preliminary injunction from one regional court in one country, it is not settled law, and Google may appeal. But it sets a direction other courts can follow.

Why it matters for AI search more broadly

The reclassification is the part with reach. If an AI summary is the platform’s own publication rather than a neutral index, the liability calculus for every AI-search product shifts, and so does the long-standing argument that these systems merely surface what already exists. That collides with the commercial logic marketers already know from generative engine optimisation: AI Overviews increasingly answer the query in place of the click, which is why so few users reach the sources.

A platform that is legally answerable for its summaries has a sharper incentive to get them right, and brands have a stronger case that being misrepresented in an AI answer is the platform’s problem to fix rather than theirs to absorb. The defensive move for marketers is unglamorous but effective: the cleaner and more unambiguous a brand’s own published information, the less room an AI has to invent a connection that was never there.

The injunction is preliminary and Google is reviewing it, so the precedent is early rather than fixed. What the Munich court has established is narrower than a global rule but sharper than commentary: when an AI summary speaks, a court has now said it speaks for the company that built it.