Digital Content Next, the publisher trade association whose members include The New York Times, asked a federal judge in New York on October 29, 2026 to accept a brief that sets Microsoft's own description of an AI "doom loop" alongside figures from its members: a 26 percent fall in unique visitors between 2023 and 2025, and a median 22 percent year-over-year drop in referred web traffic over an eight-week period in 2026.

The 5,559-word document, styled a proposed brief of amicus curiae in support of the news plaintiffs' motion for summary judgment, was docketed as Exhibit A to entry 2325 in In re: OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143, the consolidated proceedings before US District Judge Sidney H. Stein in the Southern District of New York. It joins the record only if the judge grants leave. It is advocacy from a trade body whose members include one of the plaintiffs, a fact the brief discloses, and its member figures come from surveys that, according to the brief, DCN compiled on its own and had not previously published. The brief's central claim is that the AI products at issue are eroding the advertising, subscription and licensing revenue that pays for the journalism those same products rely on.

In Short

A group representing large US publishers, including The New York Times, has asked the judge in the Times' copyright case against OpenAI and Microsoft to consider its arguments and its members' traffic numbers. It says AI chatbots answer questions using publishers' articles, so fewer people visit publishers' websites, which drains the advertising and subscription money that pays for the journalism the chatbots need in the first place - a cycle Microsoft itself called a "doom loop." The judge has not yet decided whether to accept the brief, or which side wins the case, but the outcome will help settle whether AI companies must pay for the news they use.

A phrase from Microsoft's own files

The quotation at the centre of the brief comes from Microsoft. "Our AI content strategy has started a 'doom loop' that will hurt the performance of our models and the entire web at the same time: It is highly unusual that an end-product threatens the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business with respect to its 'content supply chain,'" the passage reads, as reproduced in DCN's brief.

DCN did not unearth it. The brief cites the line to the news plaintiffs' combined summary judgment brief, which in turn cites the evidentiary record, and it names neither the author nor the date of the Microsoft document. PPC Land reported the same passage on September 20, after the plaintiffs filed the public version of their combined brief on September 17, 2026 with fewer redactions than before.

What DCN does is build an economic argument around it. The mechanism, in the brief's account, is retrieval-augmented generation, "a technological process by which, in response to a user query, the tool retrieves relevant external sources and supplies those sources to the tool so that it may generate its response." That step is part of what the brief calls grounding, generating output "from specific and identifiable external sources in response to a user's query." Because grounding pulls answers from recent, reliable reporting, DCN argues, it "is specifically designed to improve tool output by making it more similar to - and thus a direct substitute for - recent, reliable original content."

From there the loop closes. "As digital publishers decline due to Defendants' conduct, so too will the availability of reliable training data," the brief states. "That diminution will, in turn, threaten the ongoing usefulness and vitality of every LLM trained on public data." It then turns Microsoft's phrase against both defendants: "Defendants continue to pursue a course of killing the goose that lays their golden eggs."

A second admission comes from OpenAI. "OpenAI's own Head of ChatGPT put it plainly," the brief says: without traffic flowing back to content owners, "[t]here's no incentive to create content anymore." DCN does not name the executive or give the context of the remark. PPC Land's September coverage of the plaintiffs' brief identified OpenAI's Head of ChatGPT as Nick Turley.

Jason Kint, DCN's chief executive, compressed the argument into a single LinkedIn post after the filing. "DCN's amicus in NYTvOpenAI/MSFT gets at RAG economics 101: AI answers built from premium journalism may satisfy users while destroying the publishers' market (licensing/subs/ads) it depends on, putting supply chain into 'doom loop' (MSFT words)," he wrote, linking to the filing. Comments on the post were turned off.

Who is filing, and on what terms

DCN was founded in 2001 and describes itself as "the only trade organization dedicated to serving the unique and diverse needs of high-quality digital content companies that enjoy trusted, direct relationships with consumers and advertisers." According to the brief, its members have "an unduplicated audience of 259 million unique visitors and reach 95 percent of the U.S. online population." It is a non-profit 501(c)(6) business association based in New York.

The disclosures are standard. No counsel for a party authored the brief in whole or in part, and no one other than DCN paid for it, according to the filing. DCN "notes that plaintiff The New York Times is a member of DCN." Consumer Reports, whose content features in the brief's main example, is identified as a member too.

The brief is signed by Karl C. Huth of Huth Reynolds LLP in Huntington, New York, with Matthew J. Reynolds, Jennifer Sturiale (of counsel) and J. Lee Hill. Its certificate of compliance cites the court's August 7, 2026 order (Dkt. 1663) governing amicus filings. A declaration from Rande E. Price, dated October 29, 2026, was filed alongside it as Exhibit B and is the source for every DCN survey figure.

What DCN says its members lost

The brief starts from revenue mix. A June 2026 DCN member study, it says, found that in 2025 "advertising generated 84 percent of digital revenue for participating DCN member companies, while other activities, including subscriptions, generated 16 percent." Advertising on publisher sites depends on people arriving there. "Users pay for content when they visit digital publishers' webpages and view advertisements displayed on them," one of the brief's notes argues. Microsoft, by using that content upstream, "accesses the same content without ever paying for it."

The traffic figures follow. Comparing 2023 with 2025, data from participating members showed:

  • a 26 percent reduction in unique visitors to members' sites;
  • an 18 percent reduction in total views;
  • a 16 percent reduction in total visits.

The brief then cites two snapshots focused on Google. In early July 2025, DCN ran what it calls a "snap study" of members to assess the impact of Google's AI Overviews and AI Mode, both of which, according to the brief, "yield AI-generated results in response to user queries, similar to Defendants' tools here." Over the eight weeks it covered, year-over-year traffic referred to participating publishers fell at a median rate of 10 percent. DCN repeated the exercise in July 2026 for the same eight-week window. "It found that, when compared to the YoY web traffic in 2025, the YoY web traffic in 2026 had declined even more dramatically, with a median decline of 22 percent," according to the brief.

Revenue moved in the same direction, by DCN's account. "Confidential data collected from DCN's members reveal that both total advertising revenue and total overall revenue began a downward trajectory in 2025, with both measures declining 5 percent YoY," the brief states, concluding: "RAG is thus already eating away at publishers' businesses."

Two independent studies are offered as corroboration. One is a 2025 Pew Research Center analysis of data from 900 US users, which found that people who see an AI summary on Google "are less likely to click on links to other websites than users who do not see one." When PPC Land covered that study in July 2025, the headline numbers were clicks on a traditional result in 8 percent of visits with an AI summary, against 15 percent without. The other is a randomised field experiment by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University, which found that Google's AI Overviews "reduce outbound organic clicks to publishers' webpages by 39.8%." That study, which PPC Land reported on July 1, 2026, tracked 1,065 desktop Chrome users across 68,089 searches between January 7 and February 10, 2026.

Neither study measures ChatGPT or Copilot. DCN's answer is an inference: the defendants' products, it argues, present source material "that serves as a near-perfect or perfect substitute for the original source data" and "almost certainly have a similar effect."

What the figures leave out

The member data has limits the brief does not address. It does not say how many members took part in each survey, which analytics systems the figures come from, or how the 2023 to 2025 declines were separated from other changes in search and social distribution over the same period. Distribution shifted in other ways during those years. In December 2025, PPC Land reported NewzDash data on more than 400 news publishers showing traditional web search falling from 51.10 percent to 27.42 percent of their Google mobile traffic between 2023 and the fourth quarter of 2025, while Google Discover's share rose from 37.03 percent to 67.51 percent. Those figures measure shares of Google traffic rather than absolute clicks.

The brief also says the survey information "has not been previously published in any forum." That does not hold for the 2025 snapshot. On August 14, 2025, DCN published a blog post by Kint reporting a 10 percent median year-over-year decline in Google Search referral traffic to premium publishers, based on 19 member companies over eight weeks in May and June 2025. The brief presents the 2026 figure of 22 percent, the 2023 to 2025 traffic comparison and the 5 percent revenue decline as new.

Two citations are inconsistent. The brief dates the Pew study May 23, 2025, while the Pew address it gives carries July 22, 2025, the date of the report PPC Land covered. And it defines "Microsoft News Brief" twice, once as Dkt. No. 1698 filed September 4, 2026 and once as Dkt. No. 729 filed September 17, 2026.

Microsoft's data, reused

The most striking number in the brief is Microsoft's, not DCN's. Microsoft compared the click-through rate of its Bing Chat product with that of conventional Bing results, and the brief reports the outcome, citing the plaintiffs' filing: "Microsoft's data indicates that the overall click-through rate reduction between Bing Chat and Bing Web Search shows a significant reduction in users clicking to visit an external site, with as much as an 87 to 93 percent reduction for the New York Times alone." PPC Land's September coverage of the same internal comparison set out the full ranges: a reduction of 87 to 93 percent for the Times' sites, 83 to 91 percent for the Daily News sites and 51 to 94 percent for Ziff Davis sites, measured domain by domain against conventional Bing results.

DCN adds figures on how heavily the defendants' tools draw on news. "[A]pproximately 18 percent of Copilot conversations and a whopping 40 percent of ChatGPT conversations that invoke RAG rely on news-publisher content," according to the brief, citing the plaintiffs' statement of undisputed facts. The result, it argues, is that news publishers "are disproportionately harmed."

On subscriptions, the brief cites a 2025 survey of New York Times subscribers in which 36 percent of those who used or paid for ChatGPT for news agreed that ChatGPT "means I no longer need news from the New York Times at all." It also cites testimony from an executive at another newspaper that subscribers are "cancelling" because publishers are "competing with a zero price on ChatGPT, on CoPilot." The executive is not named.

And it quotes a Microsoft employee, described as the company's Director of Applied Science, who characterised the use of scraped content as "an astonishing theft of unprecedented proportions" and possibly the "largest theft of labor in human history." PPC Land's September 21 report attributed the first phrase to Brent Hecht.

One answer about Rivian

To show what substitution looks like, the brief reproduces an answer from GPT-6, which it describes as one of OpenAI's chatbot products, to the prompt "What does Consumer Reports say about Rivian's cars?"

The reply, preceded by the line "Worked for 27s," opens: "Consumer Reports' view is mixed: Rivians are fast, capable, and loved by owners, but reliability is a major weakness." It lists Consumer Reports' "biggest findings" from its 2026 rankings: first place for owner satisfaction, with 85 percent of Rivian owners saying they would buy the same vehicle again, and "last among 26 ranked brands" for reliability, with an overall predicted reliability score of 24/100, 29/100 for the R1S and 18/100 for the R1T. A table summarises Consumer Reports' road tests of the R1T pickup and the R1S three-row SUV, including 358 miles in the publication's 70-mph highway test of the Dual Max, and the answer closes by noting that coverage of the smaller R2 is still preliminary.

"GPT-6's response makes it entirely unnecessary for the user to visit the webpages - in this case, Consumer Reports, a DCN member - that served as the external sources," the brief argues, adding that substitution "makes it even less likely the user will download any given publisher's app or subscribe to the service."

The example cuts both ways. The reproduced answer carries six links labelled with Consumer Reports' initials, pointing to its Rivian overview, owner satisfaction survey, reliability rankings, R1T review, R1S review and R2 reliability page. The brief does not say how often users follow links of that kind. Microsoft's own publisher reporting does not settle the point either: the AI Performance dashboard that Bing previewed for publishers in February 2026 counts citations in Copilot and Bing's AI answers but reports no click data.

The burden of proof

The legal core of the brief is the fourth fair use factor, the effect of a use on the market for the original. DCN leans on the Second Circuit's 2024 decision in Hachette v. Internet Archive, which called that factor "undoubtedly the single most important element of fair use," and on the Supreme Court's 1994 ruling in Campbell v. Acuff-Rose, which asks courts to consider whether "unrestricted and widespread conduct of the sort engaged in by the defendant would result in a substantially adverse impact on the potential market for the original."

From those cases DCN draws two procedural points. Courts may rely on "logical inferences" of market harm rather than empirical proof, and because fair use is an affirmative defence, the defendants carry the burden of showing their use does not damage the market. "Defendants' assertion that Plaintiffs must come forward with their own 'empirical' proof of harm gets the burden backward," the brief argues.

It identifies two markets in which it says the defendants' copies act as substitutes. The first is the end user's: someone asking a question "gains access to the content without ever having to navigate to digital publishers' webpages, apps, or the like." The second is the AI company's own. Training and grounding require exact copies, DCN argues, and without scraping, "these companies would be required to invest substantial time and money to build a newsroom, hire reporters, and build trust with consumers over time, or purchase a copy of the underlying work or seek a license, like any other user of the content."

DCN also answers the defendants' argument that infringing outputs are too rare to matter. It cites Harper & Row v. Nation Enterprises for the principle that "[i]solated instances of minor infringements, when multiplied many times, become in the aggregate a major inroad on copyright," and the Third Circuit's September 29, 2026 decision in Thomson Reuters v. Ross Intelligence, which, according to the brief, rejected as insufficient the contention that the defendant had copied "only 0.08%" of the plaintiff's works. According to DCN, the plaintiffs' experts "identified at least 59,545 Copilot conversations containing verbatim content from the Asserted Works, and verbatim grounding content 'leaks' into roughly 10 to 20 percent of the Copilot outputs."

The defendants count differently. When OpenAI moved for summary judgment, PPC Land reported that its expert had found 24 instances of verbatim regurgitation in a sample of 20 million ChatGPT conversation logs, a rate of 0.00012 percent. The two figures measure different products and different things - ChatGPT outputs in one case, Copilot conversations and grounding leakage in the other - and the court will have to decide which framing governs.

Opting out is not permission, DCN argues

A large part of the brief takes aim at a Microsoft argument about control. In its summary judgment brief, Microsoft argued that "[i]t is hard to imagine a fairer use than one a copyright holder can block at will," referring to publishers' ability to keep their content out of Copilot's web grounding.

DCN calls this backwards. Under Campbell, it argues, permission is irrelevant to fair use: "If the use is otherwise fair, then no permission need be sought or granted." And Microsoft, the brief says, "is not even asking for permission to use digital publishers' copyrighted work. Instead, it is merely giving digital publishers, through the creation of meta tags, the ability to ask that their data not be used." Such a rule "would invert copyright law's default arrangement by requiring rights holders to come forward to protect their rights, rather than requiring potential users to come forward before intruding on them."

The brief does not name the tags. Bing currently treats the NoArchive and NoCache directives as grounding controls, and on October 30, the day after DCN filed, Microsoft said those directives would lose that effect on March 1, 2027, when separate preferences for AI training and AI grounding take over.

DCN also rejects the framing that publishers claim a right to traffic. "Digital publishers do not claim they are 'entitled' to web traffic," it says. "What they do claim is that Microsoft is not entitled to free, permissionless use of their content; if it wants to use that content, Microsoft needs to seek permission and 'pay' for that use."

Field v. Google and implied licences

A related defence is the implied licence: the idea that a publisher who could have blocked a crawler and did not has implicitly consented. OpenAI has made a version of that argument. In its September motion, OpenAI said it disclosed its ChatGPT-User agent in March 2023 and respected robots.txt, that the publishers did not block the agent until April 2024, and that copies taken before the block were impliedly licensed under Field v. Google.

DCN asks the court not to follow Field, a 2006 District of Nevada decision holding that an author had impliedly licensed Google to cache his pages by not using available meta-tag instructions. The brief stresses the facts of that case: the plaintiff, an author and attorney, "decided to manufacture a claim for copyright infringement against Google in the hopes of making money from Google's standard practice," created the 51 works at issue over three days and sought statutory damages above $2,500,000.

Against it, DCN sets a line of contrary authority, much of it from Judge Stein's own district and circuit. In Associated Press v. Meltwater (S.D.N.Y. 2013), the court held that "the website's failure to use the robots.txt protocol to block its access will not create an implied license," finding "no fair inference, based simply on the absence of the robots.txt protocol, that there has been a meeting of the minds." The Second Circuit in ABKCO Music v. Sagan (2022) described Field as a "more permissive test" it has never adopted. The Eleventh Circuit, in MidlevelU v. ACI Information Group (2021), declined to extend Field "regardless of whether Field was correct." And in Graham v. Prince, decided in May 2023, a Southern District of New York court found Field "neither analogous nor controlling" for images posted on social media. The case number of that decision, No. 15-CV-10160 (SHS), carries Judge Stein's own initials.

The brief adds two distinctions. Field and a 2008 Pennsylvania case that followed it concerned caching for conventional search, "a well-known, long-established use," whereas "grounding by a generative AI tool is a novelty; publishers cannot be charged with silent acquiescence to a use they had no reason to anticipate." And a search engine, in DCN's description, "act[s] as a traffic delivery vehicle getting users to the destination website," so implying a licence there "did not displace a market for actual licenses." The same cannot be said of AI output, the brief argues.

A licensing market, by DCN's account

That market, the brief says, already exists. It cites the Third Circuit's recognition in Thomson Reuters v. Ross that "the market for licensing headnotes as text to train AI is rapidly developing," the court's July 20, 2026 decision in Bartz v. Anthropic, which followed the settlement Anthropic agreed in September 2025, and the News Corp-OpenAI partnership announced on May 22, 2024.

"The existence of this market is not hypothetical," according to the brief. The defendants have licensed news and media content themselves, it says, and third parties "including Amazon, Google, Meta, Perplexity, ProRata.ai, and Mistral AI - pay to license news content for AI training and grounding." DCN says The New York Times "has licensed content to Amazon for AI-related uses, including retrieval-augmentation." The Times itself has said little about the deal's scope: when chief executive Meredith Kopit Levien discussed it in August, PPC Land noted the company had disclosed little. The brief also points to publishers' own retrieval tools, including The Washington Post's "Ask The Post AI" and the New York Post's "Hamilton" chatbot.

Microsoft is among the buyers. In October 2025, Gannett said it was partnering with Microsoft on the launch of its Publisher Content Marketplace, which Microsoft was piloting with a select group of US publishers. By the first quarter of 2026, the renamed USA TODAY Co. said AI agreements with Meta and Microsoft had a notable impact on its results. The brief's argument is that such deals "will serve to compensate digital publishers for the use of their works and offset the reductions in advertising revenue they are experiencing," and that a court-implied, royalty-free licence would undercut them.

Where the case stands

The motions DCN is weighing in on were filed on September 4, 2026, when OpenAI, Microsoft and five groups of news plaintiffs each moved for summary judgment. The plaintiffs - The New York Times Company, eight Daily News titles, Ziff Davis, the Center for Investigative Reporting and The Intercept - assert more than 10.8 million works. Both defendants requested oral argument; as of PPC Land's September coverage, no date had been set.

The record is filling with outside voices on both sides. On September 1, 2026, the United States filed a statement of interest urging Judge Stein to reject arguments that training language models on copyrighted text violates copyright law. On the same day as DCN, the Wikimedia Foundation sought leave to file a brief in support of neither party, arguing that training was likely fair use but that ChatGPT's retrieval of web pages for its answers likely was not. DCN's brief makes no such split: it treats training, grounding and output as parts of a single substitution.

Its conclusion is blunt. "Defendants' misuse of publishers' data is not 'fair', and it is not legal - and this Court should not countenance it."

Why this matters for marketers and publishers

For advertisers, the dispute is partly about supply. If DCN's figures hold, the participating members, premium sites that sell display, video and sponsorship inventory, reached 26 percent fewer unique visitors in 2025 than in 2023, and their advertising revenue fell 5 percent in 2025. A publisher whose digital revenue is 84 percent advertising, as the participating DCN members' was in 2025, has little room to absorb lost page views without cutting costs or finding new revenue such as licensing.

The brief also lays out how publishers intend to argue the referral question. They are not asking for a right to traffic, by DCN's account, but for payment when their work feeds an AI answer. That stance runs through other litigation PPC Land has followed, including Penske Media's antitrust suit against Google, filed on September 12, 2025, which alleges that Google uses publishers' content in AI Overviews without permission. Google, for its part, said in February 2026 that building controls to let publishers skip AI Overviews was a huge engineering project.

The "doom loop" line gives that position a memorable label from inside one defendant. Whether it carries legal weight is a separate matter. A remark about strategy is not a finding of market harm, and the brief's own numbers rest on member surveys whose method it does not describe. Judge Stein still has to decide whether to accept the brief at all. If he does, and if he reads the fourth factor the way DCN urges, more of the cost of the content behind AI answers would fall on the companies that build those answers rather than on the publishers that supply it.

Timeline

Summary

Who: Digital Content Next, the publisher trade association led by chief executive Jason Kint, whose members include The New York Times and Consumer Reports, represented by Huth Reynolds LLP. The brief supports the news plaintiffs and is directed against OpenAI and Microsoft, before US District Judge Sidney H. Stein.

What: A 5,559-word proposed amicus brief arguing that ChatGPT and Copilot substitute for publishers' work, that this fails the fourth fair use factor, and that publishers did not impliedly license their content by failing to opt out. It quotes Microsoft's description of a "doom loop" and adds DCN member data: a 26% fall in unique visitors from 2023 to 2025, median referral declines of 10% in 2025 and 22% in 2026 over eight-week windows, and 5% falls in advertising and total revenue in 2025.

When: Submitted on October 29, 2026, under the court's August 7, 2026 order on amicus filings, in support of summary judgment motions filed on September 4, 2026. Judge Stein has not yet ruled on whether to accept it.

Where: The US District Court for the Southern District of New York, in In re: OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143.

Why: DCN argues that AI answers built from publishers' reporting draw readers away from the sites whose advertising and subscriptions pay for that reporting, eroding the supply of reliable content the AI products depend on. For advertisers and publishers, the case will help decide whether that content is licensed for a fee or treated as fair use, and with it the economics of premium inventory on the open web.