Microsoft says Copilot cannot replace news, but its data shows 87% to 93% fewer New York Times clicks than Bing

News publishers including The New York Times on September 17 (Thursday) filed motions for summary judgment in a US federal court, with parts of the filings subsequently unredacted. The motions cite internal documents and sworn testimony from senior executives at Microsoft (Nasdaq: MSFT) and OpenAI, claiming the companies had long recognised that chatbots trained on news content and designed to provide direct answers could discourage readers from clicking through to the original news websites.

Microsoft said the remarks did not represent the company’s position and that Copilot was not a replacement for news. The allegations come from the plaintiffs’ motions and do not represent findings by the court, which has not ruled on whether either company infringed copyright.

The New York Times (NYSE: NYT) sued OpenAI and Microsoft in December 2023. Several other publishers later filed similar lawsuits, which were consolidated before US District Judge Sidney Stein in the Southern District of New York.

The figures compare two products

The motion cites representative Microsoft data comparing the rate at which users clicked through to publishers’ websites after using Copilot, formerly known as Bing Chat, with the rate after using Bing web search. The click-through rate is the percentage of link appearances that result in a click.

• The New York Times website: 87% to 93% lower than Bing web search
• Eight newspapers, including the New York Daily News: 83% to 91% lower
• Digital media group Ziff Davis, whose brands include CNET and IGN: 51% to 94% lower

The comparison is between two products from the same company, rather than a measurement of how traffic to individual websites fell over time. The publicly available filings do not state the sample size or the period covered by the data. The figures also appear only in the plaintiffs’ motions.

What three executives said

In an internal memo dated January 2023, Microsoft corporate vice-president of applied science Brent Hecht wrote that millions of people around the world would soon view large language models “scraping” everyone’s work as “an unprecedented, incredible theft”. The motion says he described it in another document as potentially “the largest labour theft in human history”.

In an internal document from January 2024, by which time the litigation was under way, Hecht raised the prospect of a “doom loop”: an end product could threaten the economic foundations of its content suppliers, ultimately harming both model performance and the web as a whole. The motion also cites Microsoft documents warning that generative AI posed a “real risk” of significantly affecting the employment of people who create training data.

The motion further alleges that OpenAI created a filter after the lawsuit was filed that suppressed only the output of the plaintiffs’ content. Hecht was concerned that this could amount to an “accidental cover-up”, making it harder for copyright owners to know which material had been used for training. OpenAI did not respond to the allegation.

A Microsoft spokesperson said Hecht’s remarks reflected the personal views of one employee, were not a legal analysis and did not represent the company’s position.

Nick Turley, head of ChatGPT at OpenAI, wrote in a June 2023 memo that publishers faced an “existential threat”. The motion also quotes him as saying that OpenAI’s products were “essentially substitutive” and would become more substitutive as they improved. Discussing ChatGPT’s browsing function, he said that once a chatbot provided an answer, users had “no good reason” to click on a link. OpenAI did not immediately respond to TheWrap’s request for comment.

In a deposition earlier this year, Microsoft chief executive Satya Nadella agreed that conversations with chatbots had “replaced” visiting original sources to obtain information. In the same testimony, he said content protected by a paywall should be licensed before being used for training or retrieval. He added that, had he known OpenAI was scraping data behind paywalls, he would have exercised his rights to require OpenAI to retrain its models. Microsoft said Nadella was discussing the broader shift in how people obtain information, which should not be conflated with the copyright issues before the court.

Other allegations in the motion

The motion alleges that an interim OpenAI training dataset contained more than 91,000 works by the plaintiffs. It also says an OpenAI researcher told president Greg Brockman about a way to bypass The New York Times’ paywall, to which he replied, “ah nice”. The motion further alleges that OpenAI systematically removed copyright-management information, such as authors’ names and copyright notices, from the plaintiffs’ works before training. It asks the court to find that the removal was deliberate and seeks statutory damages calculated separately for each article. TechCrunch contacted OpenAI and Microsoft for comment, but neither responded.

The licensing market and the legal outlook

The motion argues that a market for news licensing already exists. OpenAI has signed agreements with several media organisations, while Amazon has signed at least 23 data-services agreements. The Wall Street Journal reported in July 2025 that Amazon pays The New York Times US$20 million to US$25 million a year, although the terms of the agreement have not been made public. The Times is suing OpenAI and Microsoft while also licensing its content to another technology company.

OpenAI and media organisation Axios announced a partnership in January 2025 under which OpenAI would fund the Axios Local newsrooms and Axios would allow OpenAI to train models on content it published free of charge. The amount was not disclosed.

The limits of “fair use” remain unsettled. Two federal judges reached different conclusions on similar issues in 2025, and no US appeals court has yet ruled on the matter. The US Department of Justice filed a statement of interest in the case on September 1, arguing that training large language models on copyrighted text constituted fair use. The statement is not binding. A spokesperson for The New York Times said AI companies should pay for content under copyright law.

What to watch next

• Judge Stein’s ruling on the motions for summary judgment, including fair use, whether statutory damages should be calculated separately for each article, and whether OpenAI’s removal of copyright-management information was deliberate. No ruling date had been announced at the time of publication.
• OpenAI and Microsoft’s responses to the motions, particularly how they address the remarks by Hecht and Turley.
• Whether the sample size and statistical period for the click-through data will be disclosed in subsequent filings.