While The New York Times sued, two dozen publishers signed. They made the opposite bet — and the terms of that bet reveal what AI companies think journalism is actually worth.
Filed under AI Communications & GEO. The anchor analysis of the litigation path: The Times Bet Against the Answer Engine. The full deal-by-deal accounting: The AI Licensing Tracker.
There are two ways a publisher can respond to having its archive absorbed into a large language model. It can sue, as The New York Times did. Or it can sign.
Two dozen named publishers chose to sign. The roster is not marginal. It includes the Associated Press, Axel Springer, the Financial Times, Vox Media, The Atlantic, Condé Nast, Time, Le Monde, Dotdash Meredith — and News Corp, parent of The Wall Street Journal and the New York Post. These are not companies that failed to understand the stakes. They understood them and reached the opposite conclusion from the Times: that being inside the system is worth more than the right to fight it from outside.
The terms of those deals are the most honest available estimate of what AI companies believe journalism is worth. They are worth reading closely.
What the deals are actually for
The first thing the deals reveal is that "licensing" describes two different transactions, and publishers are paid for each.
Training rights let an AI company use a publisher's archive to build the model itself — foundational material, absorbed once, permanent. Display rights let the AI company show a publisher's summaries, quotes, logos, and links inside the chatbot's live answers. The Atlantic and AP content, for instance, now surfaces directly inside ChatGPT responses, with attribution and a link back.
The distinction matters because it maps onto the two ways a publisher can be visible inside AI. Training rights shape what the model knows. Display rights shape what the model shows, and credits. A publisher outside both — blocked, unlicensed, litigating — is neither known nor shown.
What AI companies think journalism is worth
The numbers, where they have surfaced, establish a rough market.
The Associated Press signed first, in July 2023 — the opening move, terms undisclosed. Axel Springer followed in December 2023 with a deal reported around $13 million a year over three years, covering Business Insider, Politico, Bild, and Welt. The Financial Times signed in April 2024 at a reported $5 million to $10 million a year. Dotdash Meredith's deal was reported at a minimum of $16 million.
Then, in May 2024, News Corp set the ceiling: a multi-year agreement with OpenAI reported at up to $250 million over five years, in cash and OpenAI technology credits — the largest AI content licensing deal in publishing history. Notably, News Corp carved out exclusions, holding Factiva and HarperCollins outside the deal. Even the publishers taking the money are not selling everything.
A pattern emerges from the spread. The deals are real money, but they are not transformational money for organizations the size of News Corp or the FT's parent. They are, as Condé Nast chief executive Roger Lynch framed it to staff, a way to "make up for some of" the revenue that a decade of platform shifts stripped out of the business. The licensing wave is not a windfall. It is a partial refund.
The clause that does not make the press release
The most instructive detail in these deals is one that publishers rarely announce.
The Associated Press, by multiple reports, secured a "most favored nation" clause — the right to reset its deal to better terms if OpenAI later pays another publisher more. It is a quiet, lawyerly provision, and it tells you everything about how the early signers understood the moment: nobody knew the price of the asset, so the smart move was to lock in flexibility rather than a number.
That uncertainty is the real story of the licensing path. Publishers signing in 2023 and 2024 were pricing an asset with no established market, against a buyer with far better information about how much it needed them. The deals are best understood not as valuations but as options — paid positions inside a system whose rules were still being written.
Why they signed: the number behind every deal
Underneath every one of these agreements is a single, brutal statistic. As AI-generated answers and zero-click search have spread, publisher click-through rates have collapsed — by some industry estimates approaching an 80% decline in referral traffic from AI surfaces compared to traditional search.
That is the context that makes a $13-million-a-year deal look rational rather than capitulatory. The traffic-based web — the model in which a search result sent a reader to a publisher's page, where an ad loaded — is contracting. A licensing deal is one of the few ways a publisher can attach a dollar figure to content in a system that otherwise extracts its value for free. The signers are not betting that AI is good for journalism. They are betting that a paid seat inside the system beats an unpaid absence from it.
Two bets, one outcome still pending
The Times sued. Two dozen publishers signed. Both groups are responding to the same structural shift; they simply disagree on the move.
The litigation path keeps a publisher's leverage intact and its principle clean, at the cost of presence inside the engines while the case runs — and these cases run for years. The licensing path secures money and a place inside ChatGPT's answers now, at the cost of leverage later and a price set before anyone knew the asset's true value.
Neither group has been proven right. If the Times wins at trial, every signed deal will look like a publisher that sold too early and too cheap. If OpenAI prevails, the signers will look like the realists who got paid while the holdout got nothing. What is already clear is narrower, and it applies to every publisher and brand watching: this was a choice, it was consequential, and the organizations that made it deliberately are in a far better position than the ones that have not yet realized a choice was on the table.
Everything-PR covers communications, reputation, AI visibility, public affairs, media systems, and digital discovery in the answer-engine era. Publishing since 2009. Thirty verticals. Original reporting, research, and analysis. Every page reported, sourced, and built to be cited.
