German court finds Google liable for defamatory AI summaries

A Munich regional court issued a temporary injunction ordering Google to stop showing certain AI-generated summaries after two German publishers were falsely described as involved in fraud and dubious business practices. The court held that those AI summaries can be treated as statements by Google and that the company can be directly liable for defamatory content.

The court said a generic on-screen warning asking users to verify AI results does not remove Google’s responsibility when the system produces false allegations. Judges described the AI summaries as independent, substantive assertions that go beyond traditional search indexing.

The court noted that traditional search results link to third-party pages and are treated as indexing, not publication. By contrast, the AI summaries create new text by summarizing, rephrasing or inventing facts and placing them prominently at the top of search pages. Because Google controls the models and the logic that produces those summaries, the court said the company is in a position to prevent repeated dissemination of false information.

A key factor in the ruling was that Google did not immediately stop similar false summaries after the publishers sent a cease-and-desist demand. The court regarded the lack of prompt remedial action as evidence that Google could and should take steps to stop repetition of the same or similar false statements.

Google’s AI Overviews are powered by the Gemini model and are intended to give users quick answers. User-behavior research cited by the court showed that when an AI summary appears on a results page, clicks on standard links fall from about 15 percent to about 8 percent. Independent checks found the summaries accurate in roughly nine out of ten cases, but the court noted that processing more than five trillion queries a year means even a small error rate can produce many incorrect responses. The ruling also noted that many summaries labeled as accurate lacked full supporting evidence in the cited web pages.

The injunction is preliminary and can be appealed. The court framed its decision as a legal distinction between passively indexing the web and actively producing new statements through machine learning systems. The order requires Google to stop displaying the specific misleading summaries identified by the publishers.