Traditional search results generally direct users towards information produced by identifiable third parties, leaving the user to assess and interpret those materials. By contrast, the integration of generative artificial intelligence into search engines – most notably through features such as Google’s AI Overviews at the top of search results – increasingly synthesise information into answers that may be perceived as authoritative, even where they are inaccurate or fabricated.  

Against this background, Keene v. Google LLC represents a significant developing authority concerning the potential liability of technology companies for allegedly defamatory statements produced through artificial intelligence. The litigation concerns statements generated by Google’s AI Overview concerning James Keene, including assertions that he was serving a life sentence for serious offences. The Court was required to consider whether AI generated search summaries could constitute actionable statements of fact rather than merely erroneous or otherwise non-actionable computer-generated content.

Background to the proceedings

Keene is an author and former FBI operative who served as a prison informant. In 1997, Keene pleaded guilty to conspiracy to distribute drugs and was sentenced to imprisonment. His conviction was subsequently expunged following his cooperation with federal authorities.

Between May and June 2025, Google’s AI Overview allegedly generated several statements concerning Keene in response to searches relating to his identity and net worth. Some of the outputs stated that Keene was serving a life sentence without parole, including an assertion that he had been imprisoned for the murders of three women. Another output stated that he had been convicted of drug trafficking. In attributing these claims to Wikipedia, Google’s AI Overview allegedly cited a source that did not contain the statements in question.

Keene raised complaints to Google on several occasions, asserting that the information was false and defamatory. He subsequently brought proceedings for defamation. Google sought to dismiss the claim, contending that the AI generated material should not be treated as actionable factual assertions. The United States District Court for the Northern District of Illinois considered these arguments in September 2026.

AI generated statements and defamation

The District Court assessed whether the AI Overview’s outputs could properly be characterised as statements of fact. Under Illinois law, a defamation claim requires a false statement concerning the plaintiff that is communicated to a third party and causes the requisite harm. Statements imputing criminal conduct may constitute defamation.

Google argued that their AI Overviews are inherently susceptible to error and should be understood as preliminary material intended to encourage users to conduct further research. Google also relied upon contextual matters, including that their AI overviews carry disclaimers, to argue that a reasonable reader would not regard the output as a definitive assertion of fact.

The Court rejected that reasoning at the pleading stage. In determining whether an allegedly defamatory statement is factual, the Court considered its meaning, its verifiability and the surrounding context from the perspective of an ordinary reader. Statements asserting that an identifiable person is serving a life sentence are capable of objective verification and possess a sufficiently precise meaning to be understood as factual propositions.

Significantly, the Court also rejected the proposition that providing a link to the source converts an AI generated statement into an opinion or protects the publisher from liability. A citation does not neutralise an otherwise factual assertion, particularly when the Wikipedia page did not contain the statements generated by Google’s AI Overview.

AI as an intermediary

The Court’s reasoning in Keene v. Google LLC suggests that the fallibility of artificial intelligence is not determinative of the legal analysis. Human publishers are equally capable of making mistakes, and the fact that a statement may be erroneous does not deprive the statement of its character as a statement of fact capable of giving rise to liability in defamation.

In addition, the more an AI system presents information in the form of an apparently definitive factual answer, the more difficult it may be to characterise the output as merely a neutral conduit to third party information.

At the same time, Keene v. Google LLC does not establish that every inaccurate AI generated statement will result in liability. The plaintiff must still satisfy the applicable requirements of the relevant defamation law, including falsity, publication and the necessary mental state.

Actual malice and corporate knowledge

A further significant aspect of the proceedings concerns the constitutional requirement of actual malice. Keene accepted that he was a limited public figure in relation to commentary concerning his life story. Consequently, he was required to plead facts capable of establishing that Google acted with knowledge of falsity or reckless disregard for whether the statements were false.

The Court concluded that Keene had plausibly pleaded this element. Of particular importance was his allegation that he repeatedly notified Google of the inaccurate statements and that substantially similar material continued to appear thereafter.

Therefore, in future AI related litigation, the initial AI generation of an inaccurate output may be treated differently from continued publication after a platform has received specific notice that the information is allegedly false. 

The judgment does not establish that notification automatically creates actual malice. It does, however, demonstrate that allegations concerning notice and the absence of effective corrective action may be sufficient to allow a defamation claim to have sufficient grounds.

Conclusion

Erroneous AI outputs are increasingly perceived as authoritative statements of fact and relied upon by third parties. This is reinforced by search engines’ prominent placement of AI generated summaries at the top of search results, where they purport to synthesise information from across the web into a seemingly reliable, self-contained account. The generative nature of AI means that so called “AI hallucinations” are inevitable. Put simply, AI has no inherent sense of restraint, when confronted with a gap in its knowledge, it will unavoidably generate plausible sounding content rather than admit that it does not know. When falsehoods are repeated across the web over time, it will unquestioningly reproduce them and present them as fact, and, at times, it may simply fabricate content without any apparent reason. The result may be misleading information. Where such information concerns an identifiable person or business, the resulting reputational damage may be irreparable. 

RPA offers reputation management services to monitor, counteract, and repair digital footprints that adversely affect reputation. 

RPA also provides a practical risk management response to the risks associated with AI generated information. Continuous monitoring enables potentially inaccurate or damaging mischaracterisations to be identified at the earliest opportunity. The risks are no longer merely theoretical. In the yachting and maritime industries, AI hallucinations have, in reported cases, resulted in false or unsubstantiated information entering third party vessel intelligence reports, and subsequently being relied upon by authorities in making consequential decisions, potentially including vessel detentions, arrests or other enforcement action. In such circumstances, proactive detection, verification and management are essential to mitigating the resulting reputational and legal risks.