Google Must Own Its AI Overview Errors, Court Rules

A German Court Just Reclassified AI Search

Google got sued for its search results. Not metaphorically — literally held directly liable for false claims its AI Overviews generated about two Munich-based publishers, falsely linking them to fraud schemes.

The Munich Regional Court ruled in late May 2026 that AI Overviews are not search results. They are standalone content. That distinction carries enormous legal weight: traditional search engines bear limited, indirect liability for what they surface, because they are pointing to someone else’s words. Munich said Google’s AI is producing its own words — and that changes who owns the error.

Google is appealing.

The Berlin Split

Courts do not always agree. A Berlin court reached the opposite conclusion in early June 2026, in a different case. Its ruling treated AI Overviews as another form of search output, placing Google in the same indirect-contributor category it has occupied for decades.

That is the crack in the wall Google will try to widen. Its stated position is that the Munich case involved “specific and narrow errors, not the foundational way AI Overviews displays web content.” But that framing is doing a lot of work. Google has not explained where the line sits between an error that triggers direct liability and one that does not.

The Munich-versus-Berlin split gives Google a real argument on appeal: there is no settled legal consensus, and that ambiguity tends to favor the party with the resources to keep litigating.

What the Liability Question Actually Turns On

The technical mechanics matter here. AI Overviews do not retrieve and display a publisher’s sentence. The model synthesizes across multiple sources, makes inferences, and produces new text. That generated output did not exist anywhere on the web before Google’s system assembled it — which is precisely why Munich treated it as Google’s own content.

That is not a frivolous reading. When a model hallucinates a connection between a publisher and a fraud scheme, there is no source document to point to. The error originates inside the model. Calling Google a neutral conduit in that situation is a harder case to make than it was for a hyperlink to a third-party page.

The counterargument from Google and the Berlin ruling is that the product still functions as search — users are trying to find information, not read a Google-authored article. Intent and function, not generation mechanism, define the liability frame under that view.

The Stakes Beyond This Case

A Munich appeal does not set EU-wide precedent immediately, but the direction of travel matters. If the ruling holds, Google faces a structural problem. AI Overviews run at scale. There is no manual review layer between the model’s output and the user. Direct liability for generated content at that volume would require either much tighter quality controls — which cost inference time and money — or a significant reduction in how much synthesized text the product surfaces.

Google’s current position is that the errors are an edge case, not a design flaw. That may be true statistically. It is not a legal defense.

The broader pressure lands on every company deploying AI-generated text at scale in markets where content liability rules are strict. The EU AI Act is already in motion. If German courts decide that generative output is publisher output, the compliance calculus for AI search, AI summaries, and AI-assisted news tools shifts considerably. Being the model’s operator starts to look a lot like being the editor.

The Berlin ruling buys Google time and a credible appellate argument. But the underlying question — does generation create ownership of error? — will not stay unresolved for long.