Product1 publisher3 min readPublished
OpenAI and Microsoft records describe chatbots replacing visits to the sources they draw on
OpenAI's ChatGPT chief wrote that its products "are largely substitutive, period," in internal documents made public September 17 in the Times lawsuit. That helps publishers argue over AI search, where Fast Company says the money is, but the labs still set the price.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- In an early 2023 memo, Microsoft applied science director Brent Hecht called large models "hoovering up" people's work "the largest theft of labor in human history."
- Microsoft CEO Satya Nadella testified that using chatbots "has substituted" for visiting the original sources.
- The Justice Department conceded that "an output reconstructing and disseminating an original copyrighted work may not be transformative."
- Apart from a few licensing deals, AI companies have shown little interest in building a marketplace or payment system to buy content in real time at a fair price, Pachal writes.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure OpenAI and Microsoft now have to contest their own executives' words on market harm, one of the fair-use tests the Times case will apply.
- constraint A paid market for training data is hard to build while buyers expect a ruling could make that data free to use, so archive-heavy publishers wait on the court.
- cost Publishers absorb the lost visits the executives described before any payment arrives, and the buyers decide what that payment is.
An OpenAI researcher described getting around the New York Times paywall, and OpenAI President Greg Brockman replied, "ah nice." [3] Nick Turley, OpenAI's head of ChatGPT, called chatbots an "existential threat" to publishers [4].
In Pachal's account, here is what the teams told themselves at the start: big data sets like Common Crawl already existed, and trawling the open web was how search engines had worked for decades [18]. Here is what users actually do, according to the companies' own executives. They read the chatbot's answer and stop there. Turley and Nadella both described that substitution, one in writing and one under oath [5][6].
Pete Pachal, writing in Fast Company, separates the litigation from the revenue. "Training is where the lawsuits are. Inference is where the money is," he wrote [12]. The Times sued in December 2023, when the argument was mostly about training [17]. Since then the Justice Department has filed a statement of interest saying training on publishers' content is fair use [7]. The administration argues that conceding on copyright would help China win the AI race, and President Donald Trump summed it up as "China's not doing it." [9]
The live-answer side is where the documents matter more. Pachal argues that AI search, "basically a machine for summarizing current reporting," comes close to the kind of output the DOJ said may not be transformative [19].
Pricing has not caught up. Brian Morrissey of the Rebooting suggests most content is not unique enough to command much: even if a publisher has the world's best enchilada recipe, the AI only needs one [14]. Fast Company's headline describes publishers as having buyers and "zero say in the price" [15]. The source material does not tie any deal's terms to the unsealed documents, and Pachal is cautious about the case itself. "We'll see how much this factors into the ongoing case," he wrote [16].
I think the documents help in one corner of the market. Pachal ties content's value to how unique it is and what the buyer wants to do with it [10]. Put those two variables on a grid.
- Training, interchangeable content. The DOJ backs fair use and the buyer needs one recipe. Expect a token payment or none. - Training, unique content. The leverage sits in the lawsuit, and it moves at the court's pace. - Live answers, interchangeable content. The legal footing is firmer, but the lab can answer from a rival's page, so the price falls toward the cheapest substitute. - Live answers, unique content. The substitution testimony and the DOJ's concession on outputs both point here. This is the box where a publisher has grounds to name a rate.
For a licensing lead, my recommendation is to price live-answer access separately from training and anchor the ask in the labs' own language about substitution. The tradeoff is that holding firm on price only works in that last box. Anywhere else, a refusal sends the lab to someone else's coverage. The test for each content line is whether a chatbot could give the same answer from another outlet's page. If it could, the documents improve the publisher's legal argument and leave the price where the buyer set it.
What to watch
- How the court in the Times case weighs Turley's and Nadella's substitution statements under the market-harm factor of fair use.
- Whether any AI lab builds a real-time payment or marketplace system for licensed news content.
- Whether the DOJ extends its fair-use position from training to AI search outputs.