New York Times Lawsuit: What's OpenAI's Defense?

On Thursday, a filing dropped. And only one document made the news

On September 17, in the consolidated docket that bundles news publishers' lawsuits against OpenAI and Microsoft, both sides filed for summary judgment. On one side, the brief from The New York Times and the other plaintiffs. On the other, OpenAI's brief, Microsoft's brief, and the additional statement of facts Microsoft filed the next day. Four public documents, filed in the same docket.

Within hours, four newsrooms covered the first one. Ars Technica, TechCrunch, and 404 Media on the 17th, Next.ink out of France on the 18th. Solid pieces, with accurate quotes. None of them mentioned that the other side had filed on the same day.

Nobody missed a scoop. They missed a document sitting on the same server, a few lines further down the list. We went and got it.

What these documents actually are, before we let them say anything

A summary judgment brief is a lawyer's document, not an investigative report. It explains why its client should win without ever going to trial. It picks its quotes, cuts what doesn't help, and that's the job.

The Rule 56.1 statement Microsoft filed on September 18 carries a title that's almost misleading on its own: "Statement of Undisputed Material Facts," a list of facts nobody supposedly disputes. Except the other side will dispute every one of them, point by point. The word "undisputed" works here the way a "private property" sign works on a plot of land under litigation: it announces a claim, not a ruling.

Same story on the other side. The document that made headlines Thursday is the publishers' brief, and its exhibits, the ones containing the internal documents it quotes, remain sealed. TechCrunch was the only one of the four to spell that out: much of the new information comes from The Times' own brief, not the underlying exhibits, which remain sealed.

What the defense says, and why that's not the truth either

Both defense briefs go after the same target, the one that made headlines: the 93% drop in clicks to news sites. Microsoft argues that "the Times's and Daily News Plaintiffs' experts did not even analyze web traffic," even though their entire damages theory rests on it. Its Rule 56.1 repeats that line masthead by masthead, from The New York Times to The Denver Post to The Chicago Tribune. It also claims the New York Times's conversion rate rose between November 2022 and November 2025.

OpenAI adds its own exhibit, also skipped by all four outlets: News Plaintiffs' expert did not calculate click through rates on ChatGPT and admitted that she did not "perform[] an analysis on traffic or quantifying traffic volumes in any way."

Except the plaintiffs don't deny the charge. They sidestep it legally. On page 68 of their brief, two paragraphs after that figure, they write that plaintiffs are not required to prove harm empirically: common sense suffices, citing the Hachette precedent that a rights holder doesn't need empirical data to rebut a fair use defense. In their reading, the burden of proof sits with whoever claims fair use, not with whoever is complaining.

So the two camps don't actually disagree on what was measured. They disagree on what should have been measured, and that's for Judge Sidney H. Stein to decide. That disagreement, which is the heart of the case, doesn't show up in any of the four write-ups.

The number that went viral, and what it actually measures

Everywhere this week: clicks to news sites dropped 93%. The number really is in the plaintiffs' brief, page 68, and it's defined with precision there.

It's a click-through rate per impression, compared across two surfaces: Bing Chat and Copilot on one side, regular Bing web search on the other. The brief describes "the overall click-through rate reduction between Bing Chat and Bing Web Search." This isn't a change over time, and it isn't a traffic volume.

It's the difference between an occupancy rate and a passenger count. A train can empty out row by row without the station losing a single traveler, if people are boarding the train next to it. The document backs the first claim, not the second.

The two sides agree on this specific point. Microsoft's brief describes the exact same thing: a calculation showing users "click through" to their websites at a lower rate from Copilot than they do from the traditional Bing search engine. Declic covered the same mechanism back in March, on Google's AI Overviews.

Two different ranges in the press, and nobody got it wrong

Ars Technica writes 83 to 93% for some plaintiffs, 51 to 94% for others. Next.ink writes 87 to 93% for The New York Times, 83 to 91% for the rest. The two ranges don't match, and you'd be forgiven for assuming one newsroom misread something.

They didn't. The gap is in the brief itself. Its introduction, page 2, merges the Times and the Daily News plaintiffs into a single 83-to-93% range. Its body, page 68, splits them out: "the overall click-through rate reduction between Bing Chat and Bing Web Search was 87% to 93% for The Times's websites, 83% to 91% for DNP's websites, and 51% to 94% for ZD websites."

Ars read page 2. Next.ink read page 68. Both are accurate.

That's a ten-minute check once you have the document in front of you, and a coin flip if you don't.

Only one of the four write-ups actually strays from the text. 404 Media writes that clicks "fully cratered, falling by more than 90 percent on Bing" after the scraping, and pins the number on executives who "testified under oath." The brief describes a comparison between two interfaces, not a drop over time, and attributes it to data, not testimony.

What nobody can verify, us included

The limits here need to be spelled out plainly. The expert reports are sealed: Catherine Tucker's for Microsoft, Avi Goldfarb's for OpenAI, and, above all, the deposition of the plaintiffs' expert. Every quote from that deposition comes filtered through the opponents who chose it.

The revenue and traffic charts both defenses lean on live inside those sealed reports; the briefs point to them only by figure number, without the underlying numbers. In OpenAI's filing, the Daily News and Ziff Davis traffic data is redacted. And Microsoft and OpenAI are parties here, not referees: their version isn't truer for arriving second.

What's left is a simple fact. What's been circulating since Thursday about a lawsuit that may decide the business model of online news, we received it shaped the way the side doing the suing chose to shape it. The other side had filed the same day, in the same docket. No source to protect, no document to obtain, no leak to negotiate. All it took was going to look for it.