AI Content Licensing in 2026: Who Is Actually Getting Paid, and Who Never Will

Every few weeks another AI licensing deal gets announced, and every few weeks a certain number of independent publishers read the headline and wonder when their turn comes. It’s worth being blunt about the structure of this market, because the answer for most of them is never, and the reasons are not about quality.

Start with the money that has been disclosed. News Corp’s arrangement reportedly runs around $50 million a year across the Wall Street Journal, the New York Post and its international titles. Reddit’s combined API licensing revenue from Google and OpenAI sits near $130 million annually. Wiley has disclosed more than $44 million across several separate agreements. Add up only the publicly reported figures across the whole market and you’re in the high hundreds of millions, which understates the true total, because most contracts are private and several of the biggest (Vox Media, The Atlantic, Reuters, AFP, Condé Nast, Hearst) never carried a public price.

On the buy side the concentration is stark. OpenAI leads with roughly two dozen disclosed deals, about double the next tier, where Microsoft and Meta sit at a dozen or so each. News and journalism content is the single largest category by a wide margin, with something like 48 disclosed agreements, ahead of music and audio at 16 and image and video at 12. Amazon came in later with the New York Times and then Condé Nast and Hearst. Meta ended a long silence by confirming multi-year deals with seven major publishers including CNN, People Inc. and USA Today’s owner. Anthropic, as of the middle of this year, has essentially nothing comparable, having instead agreed to a large proposed settlement in an authors’ copyright case.

The structural filter

What every one of those names has in common is a corpus large enough, fresh enough and legally clean enough to be worth the transaction cost of a negotiation. A model company doing a deal is not buying prose, it’s buying a defensible claim to a body of regularly updated text plus, increasingly, a real-time retrieval feed. Below a certain scale, the cost of papering the agreement exceeds the value of the content, no matter how good the content is.

That’s the filter, and it doesn’t move. A site producing 400 excellent articles a year is not going to be individually licensed by anyone, ever, on current market structure.

The intermediaries exist to solve exactly this problem, at a price. Startups like TollBit, ScalePost, Sphere and Defined aggregate smaller publishers into something a buyer will talk to, and take a cut of whatever comes back. Big Tech now runs marketplaces of its own. A report from the Center for Media and Digital Governance argued this leaves publishers in a double bind: hold out and get scraped anyway, join a marketplace and accept terms and take rates set entirely by the other side. Its warning was about timing. Deal structures, price precedents and governance norms established now become very hard to renegotiate later, and the window for a credible collective claim is not open indefinitely.

What smaller publishers can actually control

Nobody enjoys the answer, but it is at least in reach without a lawyer.

Selective crawler blocking rather than blanket blocking, so the highest-value material is protected while the discovery-driving material stays visible. Direct distribution, where the reader receives the work with no intermediary in between and niche newsletter sponsorships clear somewhere in the $25 to $80 CPM range. Better monetization of the traffic that still arrives, since managed header bidding typically delivers a multiple of default display revenue. And, most durably, producing the kind of work a model cannot synthesize from training data: proprietary data, expert interviews, first-person reporting, anything where the value is in the gathering rather than the phrasing.

None of that is a licensing cheque. It is, for almost everyone outside the top tier, the actual business.