Federal Judge Lets Defamation Claim Over Google AI Overviews Go Forward

A federal judge in Chicago ruled on September 14, 2026 that two Google AI Overviews falsely stating author James Keene was serving life without parole can support a defamation per se claim, rejecting Google's argument that an AI-generated search summary is not a statement of fact. The court dismissed two other statements, one as substantially true.

Information last verified on September 22, 2026. This is a developing story; we update it as the record changes.

Jurisdiction scope: This article addresses a federal trial court ruling applying Illinois defamation law and the federal constitutional actual-malice standard. It is a decision on a motion to dismiss, not a final judgment, and it binds no other court. It does not address state AI statutes or platform immunity under Section 230, which the opinion did not decide.

What Happened

James Keene is an author and television producer. In 1997 he pleaded guilty to conspiracy to distribute drugs and was sentenced to ten years. He then agreed to work with the FBI as a prison informant in the investigation of suspected serial killer Larry Hall. Hall was convicted, and Keene's own conviction was expunged. Keene wrote a memoir about the experience that became the Apple TV series Black Bird.

Between roughly May 24 and June 27, 2025, Keene alleged, at least four Google AI Overviews returned false statements about him. Three came back from searches for his net worth. One said he was "serving a life sentence without parole for multiple convictions." A second said he was "serving a life sentence without parole for the murders of three women." A third said he "was convicted of drug trafficking." A fourth, in response to a search asking whether he was still alive, said he was "serving a life sentence without parole in Butner, North Carolina."

Each of the four cited a Wikipedia page and linked to it. According to the complaint, the Wikipedia page did not contain any of the false statements.

Keene complained to Google three times: on May 27, 2025, on May 31, 2025, and again after June 27, 2025. He alleged Google apologized each time and attributed the statements to errors by its artificial intelligence. He filed a single count of defamation per se in the Circuit Court of Cook County on July 30, 2025. Google removed the case to federal court and moved to dismiss under Rule 12(b)(6).

Judge Durkin granted the motion in part and denied it in part.

What the Law Actually Says

Under Illinois law a defamation per se plaintiff must plead a false statement of fact about him, publication to a third party, and damages, and a statement imputing the commission of a crime is defamatory per se. Because Keene conceded he is a limited public figure, he also has to plead actual malice, meaning knowledge of falsity or reckless disregard for the truth. Our guide to how Illinois treats libel and slander claims covers the underlying elements in more detail.

Google made four arguments. The court worked through each.

Is an AI Overview a statement of fact? Google pointed to its standard disclaimer that AI Overviews "may include mistakes," to the linked Wikipedia source a reader could check, and to hedging language in the text. The court set the disclaimer aside because it was not identified in the complaint or the exhibits, and because Keene alleged that test searches run by his counsel produced no such disclaimer. Whether a disclaimer appeared, the court held, is a factual dispute that cannot be resolved at the pleading stage.

On the citation argument the court was blunt: "The mere act of providing a citation to an assertion does not convert that assertion from fact to opinion." It distinguished the Georgia decision in Walters v. OpenAI, where a journalist had deliberately asked a large language model to summarize a legal document, had previously received fictional answers from it, had assented to warnings about incorrect output, had reviewed multiple disclaimers that the information could be misleading or inaccurate, and had been told the model could not provide what he wanted. That context, the court said, is not this one:

"From the perspective of an ordinary user, asking an AI LLM to generate work product and receiving numerous indicia of falsity is different from a basic Google search." Keene v. Google LLC, No. 1:25-cv-11431 (N.D. Ill. Sept. 14, 2026), slip op. at 6

The court added that a user searching a name on Google "is seeking factual information, not AI work product," and that Google's own argument that the statements were objectively verifiable conceded the point it was trying to make.

Was one statement substantially true? Yes. AI Overview 3 said Keene was convicted of drug trafficking; in fact he was convicted of felony conspiracy to distribute cocaine under 21 U.S.C. section 846. The court applied the substantial truth doctrine, under which a statement is false only if it would have a different effect on the mind of the reader than the literal truth would. It noted that 18 U.S.C. section 924(c)(2) labels any felony punishable under the Controlled Substances Act a "drug trafficking crime," and that both formulations leave a reader with the same impression, that Keene was convicted of a serious drug distribution offense. The court also held that the later expungement of the conviction does not defeat substantial truth, a point worth noting for anyone reading our explainer on what record clearing does and does not erase.

Did Keene plead actual malice? The court said yes. Google argued the allegations were conclusory and that Keene never alleged he told the specific individuals responsible for publishing the overviews. The court held that allegations of repeated correspondence plausibly indicate the messages reached people at Google responsible for policing AI Overviews, and that before discovery a plaintiff will not know their names. It pointed to Google's own exhibit showing Keene submitted a legal removal request on May 25, 2025 describing his complaints in detail, and reasoned that either a person at Google received it, which supports actual knowledge, or no person reviewed the request, which supports reckless disregard.

Was everything published? Not quite. Keene conceded that the failure to allege publication of AI Overviews 3 and 4 was a clerical oversight and asked for leave to replead. The court dismissed AI Overviews 3 and 4 without prejudice, denied the motion as to AI Overviews 1 and 2, and gave Keene until October 13, 2026 to file an amended complaint.

Analysis: Why This Matters

The following is analysis from the Recording Law Editorial Team.

The interesting part of this opinion is not that an AI output can be defamatory. Courts had already assumed that much. It is the court's insistence on treating context as a fact question tied to what an ordinary user of that particular product expects.

Two defenses that the industry has treated as near-automatic came out weaker than expected here. The disclaimer defense failed not because disclaimers never work but because Google tried to assert the disclaimer's existence on a motion to dismiss, against a complaint that did not mention it and a plaintiff who said his own searches did not show it. That is a pleading-posture outcome, and it may look different at summary judgment with a record. Still, it is a reminder that a disclaimer is evidence, not a legal shield that arrives automatically.

The citation defense is the more consequential one. Google's design choice of attaching a source link to a generated summary was offered as proof that the summary is a starting point rather than an assertion. The court read it the opposite way: a factual claim with a citation attached looks more like a verifiable fact, not less. That reasoning cuts against a design pattern that is now standard across AI search products.

The actual malice holding is narrower than the headlines suggest but is the part most likely to be copied. The court did not hold that Google is liable for what its model generates. It held that repeated, documented notice, followed by continued publication of the same falsehood, plausibly alleges either knowledge or recklessness, and that the plaintiff does not have to name the recipient before discovery. The either-or framing, that a human saw the legal removal request or no human did, puts a company's own review process on both sides of the ledger. It is worth watching whether other courts adopt it.

This is a ruling on a motion to dismiss. Nothing has been proven, no damages have been awarded, and the two surviving statements still have to survive discovery and whatever comes after. The site's defamation hub and our coverage of how states are legislating around AI systems track where this line of cases goes next.

How This Affects You

Defamation law has not been rewritten. What this opinion illustrates is how existing doctrine gets applied when the publisher is a generative system rather than a person. Courts still ask whether an ordinary reader would take the statement as fact, whether the statement is substantially true, and whether the publisher had the required state of mind.

For anyone who finds a false statement about themselves in an AI-generated summary, the feature of this opinion that did the most work was the documented notice. The court leaned on a dated, detailed report to the company and the continued publication after it. That is a factual pattern, not advice about any particular situation, and whether a similar claim would survive anywhere else depends on that state's defamation law, the plaintiff's public-figure status, and the record.

This is general legal information, not legal advice. It covers a federal trial court ruling applying Illinois law and reflects sources verified on September 22, 2026. Laws change and this story is developing; consult a lawyer licensed in your jurisdiction about your specific situation.

Related articles

• Defamation laws by state: libel, slander and suing
• Illinois defamation law explained
• AI laws and regulation in the United States
• What expungement does and does not erase

---

Last updated: 2026-09-22. This is a developing story; details verified as of 2026-09-22.

Frequently Asked Questions

Did the court rule that Google is liable for defamation over AI Overviews?

No. The court ruled on a motion to dismiss, which tests only whether the complaint states a plausible claim. It denied Google's motion as to two statements, meaning the case proceeds on them. No liability has been found and no damages have been awarded.

Which statements survived and which were dismissed?

The claim proceeds on the two AI Overviews saying Keene was serving a life sentence without parole, one of which added that it was for the murders of three women. The overview calling his conviction 'drug trafficking' was dismissed as substantially true, and a fourth was dismissed because the complaint did not allege publication to a third party. Both dismissals were without prejudice.

Does a disclaimer protect an AI summary from a defamation claim?

Not automatically, and not on the record in this case. Judge Durkin declined to consider Google's disclaimer because it was not in the complaint or the exhibits and the plaintiff alleged his counsel's test searches showed no disclaimer. The court treated the disclaimer's presence as a factual dispute for a later stage.

Why did linking to Wikipedia not help Google?

The court held that the mere act of providing a citation to an assertion does not convert that assertion from fact to opinion. It reasoned that a factual statement with a citation to an outside source signals that the statement summarizes objectively verifiable facts, which supports treating it as fact rather than opinion.

How is this different from the Walters v. OpenAI case?

The court distinguished Walters on context. There, a journalist deliberately asked a large language model to summarize a legal document, had previously received fictional answers, had assented to warnings about incorrect output, had reviewed multiple disclaimers that the information could be misleading or inaccurate, and received responses saying the model could not provide the information. The court found an ordinary person running a name search on Google is in a different position.

What does actual malice mean here?

Because Keene conceded he is a limited public figure, he must plead that Google published with knowledge of falsity or reckless disregard for whether the statement was false. The court found that plausibly pleaded based on three notifications during 2025, a detailed legal removal request dated May 25, 2025, and the continued publication of similar statements afterward.

Does an expunged conviction make a statement about it false?

Not by itself, according to this opinion. The court held that the later expungement of Keene's conviction does not negate the substantial truth of a statement that he was convicted, citing Illinois appellate authority reaching the same conclusion.

What happens next in the case?

Keene was given until October 13, 2026 to file an amended complaint addressing the two dismissed statements. The claim based on the two surviving AI Overviews continues in the Northern District of Illinois before Judge Durkin.