When AI search severed the referral, it left publishers with one apparent escape route: if the click is gone, get paid for the content directly. License the archive to the AI companies that are training on it and answering with it. Stop giving the raw material away and start charging for it.

The escape route exists. It is just not open to most publishers. The disclosed licensing deals are large, and they are exclusively large publishers’ deals — News Corp at a reported $250 million-plus over five years from OpenAI and roughly $50 million a year from Meta; Reddit at $60-70 million a year; academic publishers at $10-23 million. No content-licensing deal under $10 million has been publicly disclosed.

The pattern is a winner-take-all dynamic, and it is the precise inverse of where the harm lands. The referral collapse hit the small publisher hardest — small sites lost 60% of search referrals to large publishers’ 22%. The licensing escape is available almost exclusively to the large publisher. The site that lost the most traffic is the site least able to replace it with a license.

Underneath sits an asymmetry of leverage so steep it barely qualifies as a negotiation. A single large publisher’s archive — the Wall Street Journal, the Times, the Associated Press — is a distinct, brand-name, high-trust corpus an AI company will pay to access. A single niche affiliate site’s content is one interchangeable drop in a training set the AI company can assemble without it. The large publisher has a corpus worth licensing; the small publisher has content worth scraping. Those are different bargaining positions, and the disclosed deals reflect exactly that difference.

The structural argument I want to make: the licensing market that emerged as the publisher’s answer to the referral collapse reproduces the same asymmetry it was supposed to solve — value flows to the brand-name corpus with negotiating leverage, and the long tail provides the training and grounding data for free, receiving a citation at best — which means licensing is not a path out of the collapse for the small publisher but a confirmation of it. This is the fourth piece in the Post-Wire sequence: the death of the identical paragraph (the content commoditized), the death of the referral (the channel severed), and now the asymmetry of the license (the escape route closed to the segment that needs it most).

The headline integrative finding: There is exactly one mechanism that could change this — collective licensing — and its viability is the open question of the whole piece. Individual deals are a large-publisher game by structure: leverage requires a brand-name corpus. But statutory or collective licensing — a trade-association or government-set regime that pays publishers automatically for content used, the way music royalties work — would price the long tail into the market it is currently excluded from. The honest both-sides read: the collective path is real and advancing (the News/Media Alliance’s ProRata revenue-share, Microsoft’s publisher marketplace, the UK coalition, EU and WIPO statutory-licensing proposals), but it is unproven at scale, fought hard by the platforms, and dependent on either favorable court verdicts or new law — neither of which the small publisher controls. Individual licensing confirms the asymmetry; collective licensing is the only thing that could undo it, and it is not yet built.

This essay walks the escape route and who can take it, the leverage asymmetry that decides who gets paid, the winner-take-all data, what the small publisher actually provides, the litigation track running parallel to the deals, the collective and statutory path that is the only real alternative, and the structural reading of a licensing market that reproduces the collapse it was meant to solve.

By Thorsten Meyer — May 2026

This is the fourth dispatch in the Post-Wire track — the publisher-side forensics of AI intermediation. The first walked the death of the identical paragraph; the second continued the supply-side analysis; the third walked the death of the referral. This one walks the supposed escape from the third: licensing — and finds it open to the publishers who needed it least, closed to the ones who needed it most.

The structural argument I want to make: licensing was sold to publishers as the market correcting itself — content has value, so AI companies will pay for it — but a market only pays for what has scarcity and leverage, and the small publisher’s content has neither. The large publisher’s archive is scarce (one Wall Street Journal) and carries leverage (a brand the AI company wants to be seen citing). The small publisher’s content is abundant (one of millions of niche pages) and carries no leverage (the AI company can train without any single one). The market is working exactly as a market works — and that is precisely why it does not pay the long tail.

The headline integrative finding: The licensing asymmetry is not a market failure to be fixed by better deals; it is a market success that produces an outcome the small publisher cannot live with. The only correction is to change the market’s structure — collective bargaining or statutory licensing that pays for content used regardless of any single publisher’s leverage. That correction is being attempted, on several fronts, and it is the one thing that could turn licensing from a confirmation of the collapse into a genuine escape. Whether it arrives — and whether it arrives before the small publishers it would serve have already gone dark — is the open question.

This essay walks the escape route (Section I), the leverage asymmetry (Section II), the winner-take-all data (Section III), what the small publisher provides (Section IV), the litigation track (Section V), the collective and statutory path (Section VI), and the structural reading (Section VII).

I · The escape route · licensing as the answer to the collapse

The premise crystallization. Licensing presented itself as the market’s self-correction: if AI severs the referral, the content still has value, so sell access to it directly. The premise is sound. The distribution of who can act on it is the problem.

The logic

Content for payment, replacing content for traffic: the referral was content for traffic (the prior dispatch). Licensing proposes content for payment — the AI company pays to train on and answer with the publisher’s archive, replacing the lost referral revenue with a direct fee. A clean substitution, in principle: the monetizable asset moves from the visit to the license.

The deal structures that emerged: the disclosed agreements typically combine a fixed fee for archive/training access with performance-based payments tied to how much the content is surfaced in AI responses; in exchange, publishers get attribution, links, and access to AI technology to build their own products. OpenAI alone has 18-plus publisher deals; the major platforms (OpenAI, Google, Microsoft, Meta, Amazon, Perplexity, Mistral) have each signed publisher partners.

The roster

Who has signed: News Corp (OpenAI and Meta), the Associated Press, The Atlantic, Vox Media, Axel Springer, Financial Times, Dotdash Meredith, Le Monde, The Guardian, Schibsted, The Washington Post, Reddit, Reach (Amazon), CNN, Fox News, USA Today (Meta), AFP (Mistral). The list reads like a directory of the largest media companies in the world. It is not a directory of the publishers who lost the most traffic.

The escape-route observation

Licensing is a structurally sound answer to the referral collapse — content for payment replacing content for traffic — and the roster of publishers who have taken it is a roster of the largest media companies in the world. The escape route is real. The question is not whether it works but who can walk through it — and the disclosed deals already answer that: the publishers with brand-name archives and negotiating teams, which is to say, not the long tail that the referral collapse hit hardest.

II · The leverage asymmetry · why a market pays the brand and not the tail

The mechanism crystallization. The reason licensing is a large-publisher game is not bias or oversight. It is the structure of leverage. A market pays for scarcity and leverage, and the small publisher has neither.

The scarce-corpus advantage

The large publisher has a distinct, brand-name corpus: there is one Wall Street Journal, one Associated Press, one Financial Times. Each is a distinct, high-trust, brand-name body of content an AI company has a specific reason to license — for the brand association, the real-time news, the archive depth, the trust signal in citations. Scarcity creates leverage: the AI company cannot reconstruct the Wall Street Journal from other sources, so it pays for it.

The small publisher has an interchangeable corpus: a niche affiliate site’s content — a product roundup, a how-to guide, a category page — is one of millions of functionally similar pages. The AI company can train a competent model and answer most queries without any single niche site, because the information exists in a thousand interchangeable places. Abundance destroys leverage: the AI company does not need to pay for what it can get from any of a thousand equivalent sources.

The brand-citation premium

Being cited is worth paying for — for the brand: the AI companies pay partly to be seen citing trusted sources (it improves answer credibility and reduces legal exposure). A citation of the Associated Press is worth something to the AI company; a citation of an unknown niche site is not. The same brand recognition that survived the referral collapse (the prior dispatch’s finding) is what creates licensing leverage — so the asymmetry compounds: the brands that lost the least traffic also command the only licensing fees.

The negotiation-capacity gap

Large publishers have legal and BD teams; small publishers do not: even where a small publisher might have something to license, the disclosed deals are bespoke, lawyered, multi-year contracts. The transaction cost of negotiating one exceeds what a single small publisher could ever extract. The deal structure itself — individual, negotiated, legally complex — is a fixed cost only the large publisher can amortize.

The leverage observation

Licensing pays the large publisher because a market pays for scarcity and leverage, and the large publisher’s brand-name corpus has both while the small publisher’s interchangeable content has neither. This is not a fixable flaw in how the deals are done; it is the market functioning correctly. The scarce, branded, trusted archive commands a fee; the abundant, interchangeable, unbranded page does not. And because brand recognition is exactly what survived the referral collapse, the licensing market pays precisely the publishers who were already insulated — and ignores precisely the ones who were not.

Social Work Licensing Masters Exam Guide: Comprehensive ASWB LMSW Exam Review with Full Content Review, 500+ Total Questions, and Practice Exams

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

III · The winner-take-all data · the numbers under the asymmetry

The empirical crystallization. The asymmetry is not a vibe; it is in the disclosed numbers, and the numbers describe a winner-take-all market with a hard floor below which nothing is disclosed at all.

The disclosed deals

The large-publisher numbers: News Corp’s OpenAI deal reported at up to $250 million in cash and credits over five years; News Corp’s Meta deal at roughly $50 million a year; Reddit at $60-70 million annually; academic publishers in the $10-23 million range. Meta signed CNN, Fox News, USA Today, Le Monde Group; Amazon signed Reach and (separately) a larger New York Times arrangement.

The hard floor: no content-licensing deal under $10 million has been publicly disclosed. The disclosed market begins at eight figures — which is, by itself, a statement about who the market is for. A deal sized for a small publisher would be below the threshold at which deals are even announced.

The winner-take-all shape

The market concentrates: a winner-take-all dynamic, in the analysts’ framing — the large publishers participate in (and benefit from) the emerging marketplaces and collective frameworks; the disclosed value flows to brand-name corpora; the long tail is absent from the figures. The licensing market is not a rising tide; it is a concentration of payment at the top of the publisher distribution.

The scale that dwarfs even the winners

Even the biggest deals are rounding errors to the labs: OpenAI’s reported $100 billion commitment to Nvidia for chips is roughly 200 times the annual value of its largest content-licensing deal. Anthropic’s $1.5 billion copyright settlement (September 2025) — 44% of the entire 2025 training-data market’s valuation — established that unlicensed content carries existential legal risk, but also reframed the scale: licensing is cheap relative to compute, which means the labs can afford to pay the large publishers and still treat the entire content-licensing line as a minor cost. The large publisher’s “extremely meaningful” $250 million is the lab’s rounding error.

The winner-take-all observation

The disclosed licensing market begins at $10 million, concentrates at the top of the publisher distribution, and is a rounding error to the labs even at its largest — which describes a winner-take-all market with a hard floor below which the long tail does not appear at all. The numbers confirm the leverage analysis: payment flows to the brand-name corpora, the threshold for a disclosed deal exceeds anything a small publisher could command, and the labs can comfortably pay the winners while paying the tail nothing. The market is not failing to reach the small publisher. It is structured so that it never will.

iDRM – interoperable Digital Rights Management: Interoperability Mechanisms for Open Rights Management Platforms

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

IV · What the small publisher provides · the free grounding layer

The extraction crystallization. The small publisher is not absent from the AI economy. It is present as an input — providing the training and grounding data — and absent only from the payment. That combination is the asymmetry at its sharpest.

The unpaid input

Trained on, without payment: the small publisher’s content was, in the overwhelming majority of cases, used to train the models without a license — the basis of the lawsuits (Danish publishers, Encyclopedia Britannica/Merriam-Webster, the News/Media Alliance v. Cohere, the New York Times v. OpenAI). The training already happened; the payment did not. For the long tail, the content was an input to the model with no fee attached.

Grounded on, in real time, without payment: beyond training, AI answers use real-time retrieval (RAG) to ground responses in current content — scraping the live web, including the small publisher’s pages, to construct the answer. The small publisher’s content is the grounding layer for the answer that replaces the visit. The content does double duty — training the model and grounding the answer — and is paid for neither.

The citation-not-payment substitution

The mention is the only return: where the small publisher appears at all, it appears as a citation — a named source in the AI answer, occasionally linked. The citation is offered as the value exchange. But as the prior dispatch established, the citation delivers little traffic and no payment. The small publisher’s content grounds the answer; the small publisher receives a citation; the citation pays nothing. The large publisher gets a license and a citation; the small publisher gets only the citation.

The free-substrate dynamic

The long tail is the free substrate of a paid market: the AI companies pay the large publishers for the scarce, branded corpora and take the abundant, interchangeable long tail for free as the grounding substrate. The small publisher is not outside the AI content economy; it is the unpaid foundation of it — the layer that grounds the answers the large publishers get paid to be cited in.

The provision observation

The small publisher provides the training data and the real-time grounding layer for the AI answers — for free — and receives a citation that does not pay, while the large publisher provides a scarce branded corpus and receives a license that does. This is the asymmetry at its sharpest: the long tail is not excluded from the AI economy; it is incorporated into it as the unpaid substrate. The content is valuable enough to use and abundant enough not to pay for — which is the exact definition of a commodity input, and exactly the position the first Post-Wire dispatch warned the identical paragraph was heading toward.

V · The litigation track · the leverage the small publisher does not have alone

The contestation crystallization. Running parallel to the deals is a litigation track — and it matters to the licensing asymmetry because it is the one place the small publisher’s claim has the same legal basis as the large publisher’s, even if not the same resources to press it.

The suits

The major actions: the New York Times v. OpenAI and Microsoft (ongoing); Encyclopedia Britannica and Merriam-Webster v. OpenAI (March 2026, alleging both mass-scale training copying and real-time RAG scraping, plus verbatim reproduction and trademark violation via hallucinated attribution); the News/Media Alliance members v. Cohere (13-plus publishers, 4,000 articles); the Danish DPCMO v. OpenAI; the New York Times and Chicago Tribune v. Perplexity; Penske v. Google (the referral antitrust suit from the prior dispatch).

The settlement precedent: Anthropic’s $1.5 billion copyright settlement (September 2025) — the costliest in copyright history, 44% of the 2025 training-data market’s valuation — established that unlicensed training carries existential financial risk. That precedent is the lever: it is why the labs sign licenses at all, and it is the basis on which the News/Media Alliance’s Danielle Coffey predicts that favorable verdicts will create “an expectation of payment” and a functional marketplace.

Why litigation is also asymmetric

The same resource gap: litigation, like negotiation, favors the publisher with legal resources. The New York Times can sustain a multi-year case against OpenAI; a niche affiliate site cannot. The legal claim may be identical — unlicensed use of copyrighted content — but the capacity to press it is not. The small publisher’s strongest theoretical leverage (the copyright claim) is the one it can least afford to exercise alone.

The collective-litigation exception

Where the small publisher has a chance: the class or the association: the News/Media Alliance suit against Cohere is the model — a trade association aggregating many publishers’ claims into a single action with shared legal resources. Collective litigation is the small publisher’s only practical access to the litigation leverage that produces licensing deals. The same logic, foreshadowing the next section, applies to licensing itself: aggregation is the small publisher’s only route to leverage.

The litigation observation

The litigation track is the legal basis for licensing — the Anthropic settlement and the pending suits are why the labs pay at all — but it is as asymmetric as the deals, favoring the publisher who can afford a multi-year case, except where publishers aggregate into collective actions. The copyright claim is the one piece of leverage the small publisher shares equally with the large publisher in principle; the capacity to press it is what differs. And the only way the small publisher accesses that leverage is the same mechanism that could fix the licensing asymmetry: collective action.

VI · The collective and statutory path · the only real alternative

The correction crystallization. Individual licensing is, by structure, a large-publisher game. The only thing that could price the long tail into the market is a mechanism that pays for content used regardless of any single publisher’s leverage — collective bargaining or statutory licensing. This is the open question of the whole piece.

The collective-marketplace approach

ProRata and the revenue-share model: the News/Media Alliance struck an agreement with ProRata, where member publishers can license content into the Gist.ai product on a 50% revenue-share model. This is the long tail’s most direct shot — aggregate many publishers, share revenue by usage, lower the per-publisher transaction cost below the prohibitive threshold of an individual deal. Aggregation converts a thousand un-negotiable small claims into one negotiable collective one.

Microsoft’s publisher marketplace: Microsoft launched a marketplace to broker licensing between publishers and AI developers — publishers post content and terms, developers license what they need, Microsoft handles the infrastructure (and takes a cut). A marketplace lowers the transaction cost of a deal, which is the fixed cost that excluded the small publisher — in principle, opening participation below the $10 million floor.

The statutory-licensing approach

The government-set regime: policymakers in Europe, Indonesia, Latin America, and at WIPO are exploring statutory licensing — a legal requirement that AI companies pay publishers automatically for content used, with pricing set by regime rather than negotiation, modeled on how music royalties work. Statutory licensing is the only mechanism that pays the long tail by right rather than by leverage — it removes the negotiation the small publisher cannot win and replaces it with an automatic payment.

The European tilt and the Brussels Effect: Europe’s longer tradition of collecting societies and licensing regimes makes statutory licensing more achievable there; the EU publishers’ antitrust complaint against Google, Spain’s €524 million Meta fine, and the UK publisher coalition all point the same direction. If Europe sets a statutory regime, the Brussels Effect could propagate it — which would matter most to exactly the kind of European niche-publisher operation that the individual-deal market ignores.

Why the path is uncertain

Unproven, fought, and slow: collective and statutory licensing are real and advancing, but none is proven at scale. The platforms fought and weakened earlier bargaining-code laws (Australia’s News Media Bargaining Code) all over the world. Statutory regimes depend on new law or favorable verdicts — neither fast, neither controlled by the publisher. There is still no standardized model for pricing news content. The collective path is the only real alternative, and it is the one least within the small publisher’s power to guarantee.

The collective observation

Collective bargaining and statutory licensing are the only mechanisms that could price the long tail into the licensing market — by aggregating un-negotiable small claims into negotiable collective ones, or by replacing leverage-based negotiation with payment by right — and both are real, advancing, and unproven. ProRata’s revenue-share, Microsoft’s marketplace, and the European statutory proposals are the small publisher’s actual hope. But they depend on aggregation holding, marketplaces reaching below the floor, courts delivering verdicts, or legislatures passing law — and the small publisher controls none of those, which means its escape from the licensing asymmetry depends on a correction it cannot itself build.

What this is not

It is not a claim that licensing is a scam. For the large publisher, the deals are real and meaningful revenue. The asymmetry is in who can access them, not in whether they are genuine.

It is not a claim that the labs are uniquely predatory. The licensing market is functioning as a market — paying for scarcity and leverage. The outcome is a market outcome, not a conspiracy.

It is not a claim that collective licensing will fail. It might well succeed, especially in Europe. The claim is only that it is unproven and outside the small publisher’s control — a hope, not yet a fact.

The synthesis observation

The licensing market that emerged as the publisher’s answer to the referral collapse reproduces the same asymmetry it was meant to solve: value flows to the brand-name corpus with scarcity and leverage, while the long tail provides the training and grounding data for free and receives a citation that does not pay. Individual licensing is a large-publisher game by structure, not by accident; the disclosed market begins at $10 million and concentrates at the top; the small publisher is the unpaid substrate of a paid market; and the only correction — collective or statutory licensing — is real, advancing, and not within the small publisher’s power to guarantee.

There is no single answer. Anyone offering one is selling something. What is unambiguous is that licensing, sold as the escape from the referral collapse, is open to the publishers who were already insulated from that collapse and closed to the ones who were not. The content has value; the small publisher’s content has value; but a market pays for leverage, and the long tail has none — so the license that saved the Wall Street Journal does not reach the niche site, and the only thing that could is a market the small publisher cannot build alone. The escape route is real. For most of the publishers who needed it, it leads to a door they cannot open.

That is the structural editorial question the license sits on top of. It is a market working correctly toward an outcome the long tail cannot survive. It is value flowing to the brands that lost the least and away from the publishers that lost the most. It is the small publisher as the free substrate of a paid economy. And it is the layer where the future of independent publishing is decided not by whether the content is worth paying for — it is — but by whether the small publisher can aggregate into the only kind of leverage that gets paid, before the licensing market finishes confirming the collapse the referral piece described.

About the Author

Thorsten Meyer is a Munich-based futurist, post-labor economist, and recipient of OpenAI’s 10 Billion Token Award. He spent two decades managing €1B+ portfolios in enterprise ICT before deciding that writing about the transition was more useful than managing quarterly slides through it. He runs StrongMocha News Group, a network of more than 450 niche WordPress magazines built on the DojoClaw editorial engine. More at ThorstenMeyerAI.com.

Related Reading · the Post-Wire track

This dispatch

• This piece · The license · the licensing-asymmetry forensic — how the AI content-licensing market reproduces the referral collapse it was meant to solve, paying the brand-name corpus and stranding the long tail · empirical-clay dominant, structural-slate and labor-rose balance

The track

• The referral · Post-Wire 03 · the collapse this licensing market was the supposed escape from — the click economy becoming a citation economy
• The death of the identical paragraph · Post-Wire 01 · the supply-side commoditization that left the small publisher’s content interchangeable — and therefore un-licensable
• Post-Wire 02 · Post-Wire 02 · the continuing publisher-side intermediation analysis
• Forthcoming · The generative-engine-optimization shift · what it takes to be the cited source rather than the licensed one, and whether GEO is a durable discipline or a temporary arbitrage · transition-bronze register

Adjacent tracks

• The runway · Enterprise Reorg 04 · the labs whose licensing budgets are rounding errors against their compute commitments
• The deployment · Enterprise Reorg 03 · the same labs vertically integrating into the services layer while content licensing stays a minor line
• The mandate · Agentic Commerce 03 · the European regulatory architecture where statutory licensing is most achievable

Sources

The licensing roster and structure

• Digiday · timeline of major publisher–AI deals 2025 — the deal roster (Axios, AP, Guardian, Schibsted, Washington Post, AFP/Mistral) · fixed-fee-plus-attribution structures · the News/Media Alliance + ProRata 50% revenue-share model · OpenAI funding four Axios local newsrooms · digiday.com
• Digiday · Scorecard: publishers rate Big Tech’s AI licensing — OpenAI’s 18 deals, “docked for not returning publisher (big and small) calls” · Amazon’s NYT deal (paid without AI-search cannibalization) · Microsoft’s 8 partners + Julia Beizer hire · Google “careful not to describe these as new licensing deals” · digiday.com
• A Media Operator · key questions around OpenAI’s licensing deals — the roster (News Corp, Atlantic, Vox, Axel Springer, FT, Dotdash Meredith, Le Monde, AP) · the fixed-fee-plus-performance structure · “absolve them of the theft / settling without litigation” (Lessin) vs “take the money and play ball” (Blodget); FT’s Ridding “pulling up the drawbridge is not a strategy” · amediaoperator.com
• MediaCopilot · Microsoft launches AI-licensing marketplace — the bilateral-to-marketplace shift · publishers post content and terms, Microsoft takes a cut · Meta’s December 2025 deals (CNN, Fox News, USA Today, Le Monde Group) · model-churn complicating system-specific deals · mediacopilot.ai

The winner-take-all numbers

• Will Scott · how AI licensing deals determine search visibility — the core finding: “Major deals exclusively involve large publishers: News Corp ($250+M over 5yr), Reddit ($60-70M annually), academic publishers ($10-23M). No deals under $10 million have been publicly disclosed, creating a winner-take-all dynamic.” · OpenAI’s ~$100B Nvidia commitment ≈ 200x its largest licensing deal · Anthropic’s $1.5B settlement = 44% of the 2025 training-data market · 60% of major news sites block AI crawlers · willscott.me
• Playwire · News Corp $50M Meta deal / Danish publishers sue — News Corp’s ~$50M/yr Meta deal (WSJ, NY Post, Times of London, The Australian) + the reported $250M/5yr OpenAI deal · Reach–Amazon (Mirror, Express) · “smaller publishers face tougher choices with limited legal resources and less negotiating power… collective bargaining through industry groups is emerging as a viable alternative” · playwire.com
• BuzzStream · news publishers with AI partnerships 2026 — “partnerships don’t guarantee ChatGPT citations” · Perplexity signs the most and is blocked the most · “no AI company clearly explains how licensed content is used” · the per-platform partner counts · buzzstream.com

The litigation track

• Press Gazette · news AI deals and lawsuits — News Corp–Meta and Reach–Amazon (usage-based comp) · Encyclopedia Britannica & Merriam-Webster v. OpenAI (training + RAG scraping + verbatim reproduction + hallucinated-attribution trademark) · Danish DPCMO v. OpenAI (refused mediation) · News/Media Alliance members v. Cohere (4,000 articles) · pressgazette.co.uk

The collective and statutory path

• Poynter · a new global push to make AI pay for news — statutory-licensing proposals in Europe, Indonesia, Latin America, WIPO · the failed Australian News Media Bargaining Code precedent (platforms “fought back and weakened or blocked similar laws”) · News/Media Alliance’s Coffey: favorable verdicts → “expectation of payment” and “a functional marketplace” · Europe’s collecting-society tradition and the Brussels Effect · “no standardized model for pricing news content” · individual deals “help some outlets but may do little for the broader news ecosystem” · poynter.org

The Post-Wire track backbone

• The referral / The death of the identical paragraph · Thorsten Meyer · Post-Wire 03 / 01 · the collapse this licensing market was the escape from, and the commoditization that left the small publisher’s content un-licensable · the licensing asymmetry is the third confirmation: content commoditized, channel severed, escape closed

Key reference figures crystallized

• The escape route: content for payment replacing content for traffic · fixed fee + performance + attribution + tech access · OpenAI 18+ deals; all major platforms signing
• The roster: News Corp, AP, Atlantic, Vox, Axel Springer, FT, Dotdash Meredith, Le Monde, Guardian, Schibsted, Washington Post, Reddit, Reach, CNN, Fox, USA Today, AFP — a directory of the largest media companies
• The winner-take-all data: News Corp $250M+/5yr (OpenAI) + ~$50M/yr (Meta) · Reddit $60-70M/yr · academic $10-23M · no disclosed deal under $10M · OpenAI’s ~$100B Nvidia commitment ≈ 200x its largest licensing deal · Anthropic $1.5B settlement = 44% of 2025 training-data market
• The leverage asymmetry: scarce branded corpus (leverage) vs interchangeable abundant content (none) · brand-citation premium · the negotiation-capacity gap · the same brand recognition that survived the referral collapse creates the licensing leverage
• What the small publisher provides: training data + real-time RAG grounding, for free · citation that does not pay · the unpaid substrate of a paid market
• The litigation track: NYT v. OpenAI/Microsoft, Britannica/Merriam-Webster v. OpenAI, NMA v. Cohere, Danish DPCMO v. OpenAI, NYT/Chicago Tribune v. Perplexity · Anthropic $1.5B settlement as the leverage precedent · collective action as the small publisher’s only access
• The collective/statutory path: ProRata 50% revenue-share (NMA) · Microsoft marketplace · EU/Indonesia/Latin America/WIPO statutory licensing · the failed Australian bargaining code · the Brussels Effect · aggregation as the only route to long-tail leverage