Which Publishers Have AI Licensing Deals, and What Are They Worth?

What we counted

We took a single trade timeline of the major publisher–AI company deals announced during 2025, and recorded for each entry the date it was announced, the publisher or publisher group involved, the AI company on the other side, and whatever the timeline stated about financial terms. Where an entry covered several titles at once we recorded it as a single announcement and noted the scope separately.

Sixteen entries met that description. They run from mid-January, with Axios–OpenAI and Associated Press–Google, through to December, with a seven-publisher Meta round and a multi-publisher Google round.

We counted announcements rather than contracts. Several entries cover groups rather than single titles — one Prorata entry covers more than 500 publishers, and the USA Today entry covers a network of around 200 local publications together with its sports wires — so the number of publishers operating under licence is very much larger than the number of announcements.

The AI companies involved span the field: OpenAI, Google, Amazon, Microsoft, Meta, Mistral, Perplexity and Prorata all appear. That breadth is itself a change from the period when OpenAI was effectively the only buyer.

Announcement dates are not contract dates, and a timeline compiled by one publication is not a census of the market. Deals struck quietly, renewed without announcement, or reported only in a local market will not appear here at all. We treat this as a well-documented sample of the year's visible deal-making rather than as a complete register.

The terms column is empty

The finding that survives every cross-check is an absence. Across all sixteen announcements recorded in the timeline, financial terms were disclosed in none of them. The recurring entry in that column is the phrase “not disclosed”, repeated for every deal in the year regardless of the size of the publisher or the buyer.

This is not an artefact of one publication's reporting. Coverage of Meta's December round states the position directly: “All companies are being tight-lipped on the terms of the deals, which remain undisclosed.”

The single exception to total opacity is structural rather than numerical. A News/Media Alliance arrangement with Prorata was announced on a stated 50% revenue-share model — a rate rather than a sum, which at least allows a publisher to reason about proportion even when the underlying revenue remains entirely unknown.

An empty terms column across an entire year of deal-making is unusual even by the standards of private commercial agreements, and its consequence is concrete: there is no public benchmark for what a publisher archive is worth, no per-article or per-year rate anyone can cite, and no way for a smaller publisher to know whether an offer it receives is reasonable.

It also means every confident market total in circulation is built on estimates. There is no disclosed base to aggregate.

What the circulating numbers actually are

Two figures dominate discussion of this market. The New York Times–Amazon agreement has been reported at $20 to $25 million per year, covering the Times, NYT Cooking and The Athletic. The News Corp–OpenAI arrangement has been reported at more than $250 million across five years.

Both are reported figures attributed to sources, not terms published by the parties. We flag them as such for the same reason we flagged the News Corp number when it appeared in our earlier report: in this market the distinction between a reported estimate and a disclosed term is the whole of the reliability question.

Aggregator sites publish considerably more specific material — total deal counts, per-outlet values, citation-uplift percentages expressed to a decimal place. We did not use any of it.

Those figures are not traceable to filings or to statements by the parties, and several of the sites publishing them sell services whose value those very figures support. That is a conflict worth naming, and it applies to us as much as to them — we sell placement, which is why this report restricts itself to numbers a reader can follow to a source and check independently.

Where a number appears here, it appears with its status attached. That is less satisfying than a market total, and it is what the evidence supports.

Why the opacity matters to anyone buying coverage

The commercial logic of these deals is that publisher archives have value as training and retrieval material. That value is being priced privately, which means nobody outside the negotiations can say what editorial content is worth to an AI company.

For a company paying for placement the relevant inference is narrow but real. The outlets signing these agreements are being paid for content their own journalists produced, and a placed article sits inside the same licensed corpus without the placing company being party to the arrangement or receiving any share of what the archive earns.

That is not an argument against placement. It is an argument for understanding what is being sold to whom: the publisher monetises the archive twice, once to the advertiser and once to the model developer.

It also bears on crawler policy. Publishers restricting AI crawlers while licensing to named partners are drawing a commercial boundary rather than a technical one, which is the pattern we found when examining which crawlers publishers actually name.

The practical consequence for a buyer is that access and permission are now negotiated per-company, so where a given article ends up depends on deals the buyer cannot see.

What would make this measurable

Three things would turn this from an anecdote into a market. Disclosed terms in at least a meaningful subset of deals; a consistent unit of account, whether that is per-article, per-token, per-year or a share of revenue; and public confirmation from both parties rather than figures attributed to people familiar with the matter.

None of the three currently exists. Until they do, the only honest description of AI licensing economics is directional: more buyers than last year, broader coverage, and prices nobody outside the room can state.

A partial route exists for public companies, whose filings occasionally disclose material contract revenue in aggregate. That is worth watching, though it typically arrives late and without per-deal detail.

We would revisit this count if a trade body or regulator published a systematic register, or if a meaningful number of parties began disclosing terms.

Until then, anyone citing a total for this market is citing an estimate, and should say so.

What the direction of travel already shows

Even without disclosed terms, the shape of the year is legible. Deal-making opened with OpenAI and Google and closed with first-time agreements from Microsoft and Meta, which means the buyer side broadened from one dominant purchaser to most of the major model developers.

The seller side broadened further and faster. Group arrangements now cover hundreds of titles at once, so the practical question for a mid-sized publisher has shifted from whether a deal is available to which consortium to join.

Structure is diversifying alongside volume. A stated revenue share sits beside flat annual fees and multi-year archive licences, which suggests the market has not yet settled on a unit of account.

What has not changed is disclosure. A year that produced sixteen visible announcements produced no published terms, and the two figures in circulation remain sourced reporting rather than confirmed sums.

For anyone commissioning coverage, the durable inference is that the onward destination of a published article is now governed by commercial agreements invisible to the party who paid for it. The placement is visible, the licence is not, and no amount of diligence on the outlet reveals which models will ingest the result.

Methodology and limitations

On 19 September 2026 we took Digiday's timeline of major publisher–AI company deals announced in 2025 (published 1 January 2026, by Sara Guaglione) and recorded every entry with its date, publisher, AI company and stated financial terms. Sixteen entries qualified. We corroborated the December Meta round against Digiday's separate report of 5 December 2025 by Jessica Davies and Sara Guaglione, and took the reported New York Times–Amazon range from Digiday's subsequent coverage. This is a count of announcements in one trade source, not a census of the market: deals not covered by that timeline are not represented, group announcements are counted once, and announcement dates are not contract dates. We deliberately excluded figures published by SEO and vendor aggregator sites — including totals, per-deal values and citation-uplift claims — because they are not traceable to the parties or to filings. No financial term in this report is presented as confirmed.

Conclusion

Sixteen publisher–AI deals were announced in 2025 and not one came with disclosed financial terms. That is the finding. The two sums in general circulation — roughly $20–25 million a year for the New York Times from Amazon, and more than $250 million over five years for News Corp from OpenAI — are reported estimates, and should be described that way every time they are used. A market this active with a terms column this empty is one where nobody outside the deals can say what editorial content is worth, and anyone quoting a confident total is quoting something they cannot source.

Frequently Asked Questions

Sources

1. trade press Sara Guaglione, “A timeline of the major deals between publishers and AI tech companies in 2025”, Digiday, 1 January 2026. Primary basis for the count of sixteen announcements and the absence of disclosed terms.
2. trade press Jessica Davies and Sara Guaglione, “Meta enters AI licensing fray”, Digiday, 5 December 2025. Source of the quoted statement that terms remain undisclosed.
3. trade press Digiday coverage of the New York Times–Amazon agreement, 2025, cited for the reported range of $20–25 million per year. Reported, not confirmed.