OpenAI and Microsoft have asked a federal judge to throw out the biggest media copyright case of the decade — and their argument comes down to arithmetic.
The motion for summary judgment was filed on September 4 in the Southern District of New York, before Judge Sidney H. Stein, in the consolidated multidistrict litigation numbered 25-md-3143. It is the first filing that pulls the defendants’ whole evidentiary story into one place. Two days earlier, the US Department of Justice had filed a statement of interest backing OpenAI’s fair use defence — an unusual move for a non-party in a private dispute.
The number that anchors the whole defence
Plaintiffs assert 10.8 million articles. OpenAI was ordered to produce 20 million ChatGPT conversations in discovery. Across that corpus, its experts say they found 24 instances of verbatim reproduction, the longest running 29 and 43 words.
That works out to a reproduction rate the brief puts at 0.00012 percent, with the broader range calculated between 0.00011 and 0.000002 percent — figures OpenAI says line up with the publishers’ own numbers.
The plaintiff bloc is not small. It includes The New York Times Company (asserting 6,030,928 articles), the Daily News group (Chicago Tribune, Orlando Sentinel, Sun-Sentinel, Denver Post, Mercury News and others), Ziff Davis (IGN, CNET, ZDNET, PCMag, Mashable, Everyday Health), the Center for Investigative Reporting and The Intercept.
Three defence lines beyond the counting
- Browse is really search. OpenAI says more than 96 percent of the claimed Browse infringement involved Bing snippets only, with no request ever hitting the publisher’s server. If accepted, much of the case becomes an argument about search snippets.
- Implied licence by robots.txt timing. OpenAI disclosed the ChatGPT-User agent in March 2023 and said it honoured robots.txt. The publishers did not block it until April 2024 — about a year later, by which point 48 percent of major news sites had already blocked some AI crawler.
- No market harm. The brief points at the Times passing 12 million subscribers in 2025, targeting 15 million by 2027, with Q2 2026 digital ad revenue up 20.7 percent to $114.0 million.
The publishers’ sharper counter: dilution, not copying
Publishers argue harm is legally beside the point here, because fair use is an affirmative defence and the burden of proving no harm sits with OpenAI. They cite copying at acquisition (WebText, WebText2, Common Crawl, plus the 1.8-million-article New York Times Annotated Corpus licensed for non-commercial use), and custom GPTs with names like Bypass Paywall and Remove Paywall.
Their dilution argument is the one marketers should read twice. They point to Prism News — roughly 200 AI-generated publications posing as local newsrooms, staffed by four people total — and to an Integral Ad Science forecast that up to 90 percent of web content could be machine-generated by 2026, with single sites producing 1,200 articles a day. Generating a million 500-word articles costs about $6,800. The Times spends close to $2 billion a year for roughly half a million works.
Why this matters for your traffic model
The economics of substitution do not require verbatim copying. And the referral data is already moving: OpenAI’s crawl-to-referral ratio hit 1,500 to 1 by June 2025; Google’s went from 2 to 1 in 2015 to 18 to 1. Separate research measured AI Overviews cutting publisher clicks by 39.8 percent.
Practical read for growth and content teams:
Stop modelling organic search as a stable channel. Build owned-audience assets — email, community, app — that survive a click-free SERP. Audit your crawler policy deliberately, because the timing of your robots.txt decisions is now a litigated fact pattern. And differentiate on things a $0.0068-per-article content farm cannot fake: proprietary data, original reporting, named expertise.
Two precedents frame the outcome: Meta won summary judgment on fair use in June 2025, and Anthropic settled the authors’ class action for $1.5 billion in September 2025 — the largest copyright settlement on record, which set a price without setting a rule. Both defendants have requested oral argument. No date has been set.
Source: PPC Land
