News Corp's licensing arrangement reportedly averages around $50 million per year across its portfolio of the Wall Street Journal, New York Post, and its global titles. Wiley has disclosed more than $44 million across multiple separate AI licensing agreements. Reddit's combined API licensing revenue from Google and OpenAI sits near $130 million a year. These are the headline numbers that generate coverage every time a new deal is announced, and they create a reasonable but mistaken impression among smaller publishers that a similar deal might eventually be within reach for them too.
David Skok, CEO of The Logic, made the more uncomfortable structural argument directly in Nieman Journalism Lab's analysis: as long as participation in a search engine's index implicitly means participation in that same company's AI training pipeline, the price for that data is effectively zero, because the alternative to accepting the implicit bundling is total exclusion from search visibility altogether. Most publishers cannot afford that trade. Separately, StrongMocha's 2026 analysis of the licensing landscape reached the same conclusion from a different angle: a market pays for scarcity and negotiating leverage, and a small publisher structurally has neither. There is one Wall Street Journal. There is one Associated Press. An AI company cannot reconstruct their specific reporting from any other source, so it pays for exclusive or near-exclusive access. A smaller publisher's content, by contrast, usually overlaps substantially with dozens of other sources covering the same topics, and that redundancy is precisely what strips away any negotiating leverage.
What Does the Actual AI Content Licensing Market Look Like in 2026?
Analysis tracking disclosed licensing deals found news and journalism content accounts for the single largest category by a wide margin, 48 deals in one comprehensive tracker, more than half of all disclosed agreements, ahead of music and audio at 16 deals and image and video content at 12. On the buy side, OpenAI leads with 24 disclosed deals, roughly double the next tier occupied by Microsoft and Meta at approximately 12 each, a concentration that reflects a deliberate strategy of locking in fresh, regularly-updated content sources earlier and at greater volume than competitors, rather than an even, industry-wide willingness to pay.
The market has also visibly shifted in structure since the earliest 2023 and 2024 deals. Early agreements were largely one-time archive purchases: a fixed sum for access to a publisher's historical back-catalogue for training purposes. The more recent 2026 pattern, most clearly exemplified by Reddit's arrangement and by January 2026's formalised Wikimedia Enterprise partnership with Amazon, Meta, Microsoft, Perplexity, and Mistral, is a shift from selling a static archive to renting a continuously refreshed, real-time feed, treating the data less like a one-time asset sale and more like an ongoing subscription product. This shift matters directly for AEO strategy, because it signals that AI companies increasingly value freshly published, regularly updated content specifically for its currency, not just its historical completeness, reinforcing the same freshness principle covered throughout the content decay research elsewhere in this series, now visible at the level of formal commercial data contracts rather than just citation algorithm behaviour.
Why Does the Deal Structure Systematically Exclude Smaller Publishers?
Three structural factors, consistent across every serious 2026 analysis of this market, explain why the headline numbers above are not a realistic template for most publishers to follow.
Scarcity is the entire basis of the market, and most content is not scarce. An AI company pays for content it genuinely cannot reconstruct or approximate from other freely available sources. Original investigative reporting, a globally unique archive, or a uniquely authoritative reference work all qualify. A locally-focused blog, a mid-size trade publication, or a general news aggregator covering the same stories as dozens of other outlets does not carry the same scarcity value, regardless of the quality of the writing itself, because the AI company can obtain functionally equivalent coverage of the same underlying facts elsewhere at no cost.
The search-training bundling problem removes the negotiating leverage that would otherwise exist. As Skok's analysis lays out, when a company's search crawler and its AI-training crawler are effectively the same infrastructure, or governed by intertwined policies, a publisher cannot cleanly separate "I want to be found in search" from "I am consenting to AI training use." Blocking one, in practice, blocks both, and for the overwhelming majority of publishers dependent on search-referred traffic for their core business, that trade is untenable. Without the ability to credibly withhold access, there is no price discovery and no competitive tension driving a licensing fee upward.
Even where a formal rights-reservation mechanism exists, its practical enforcement remains weak. The EU's Text and Data Mining exception, and the associated opt-out mechanism it provides for rights holders, has already run into documented enforcement difficulty in early court decisions, including the LAION case before the Hamburg District Court and the DPG Media case in Amsterdam, both of which surfaced real-world problems with non-standardised, technically unsuitable opt-out signalling, robots.txt among them, being relied upon as if it constituted a clear, enforceable legal reservation of rights when its actual technical and legal force is considerably murkier than that.
What Should a Small or Mid-Size Publisher Actually Do About This?
Given that realistic path to direct licensing revenue is closed to most publishers, at least under the current market structure, the more productive strategic question shifts from "how do we get paid for AI training access" to "how do we maximise the citation and traffic value we can still control directly." Three concrete actions follow from that reframing.
Treat robots.txt and llms.txt as visibility-shaping tools, not licensing leverage. Given that blocking AI crawlers largely means forfeiting AI citation visibility entirely, for most publishers the more strategically sound default is allowing the crawlers that drive citation and referral traffic, as covered in full in the AI crawlers guide, while using a well-structured llms.txt file to actively shape which of your content gets prioritised for retrieval and citation, rather than leaving that entirely to chance. This does not generate licensing revenue, but it does maximise the citation and referral traffic value available from the access you are granting regardless.
Invest directly in the parts of the business that remain fully within your own control. Skok's Nieman Lab analysis is explicit on this point: publishers adapting fastest to a licensing market that is not materialising broadly are the ones redirecting investment toward owned, first-party assets, direct audience relationships through newsletters and memberships, proprietary first-party data, live events, and increasingly, structured content specifically engineered for the agentic AI systems covered in the agentic AI search guide elsewhere in this series, positioning for a future where those systems may eventually require licensed, structured inputs even if the current bulk-training market does not pay for them.
Pursue collective or platform-mediated revenue-share arrangements rather than individual licensing negotiations. Individual small publishers lack negotiating leverage. Aggregated participation in a platform-mediated revenue-share programme changes that calculus somewhat. Perplexity's Comet Plus programme, distributing a defined share of subscription revenue to participating publishers based on citation frequency, and ProRata's model of paying out fully half its Gist.AI product revenue to publishers, both represent a meaningfully different and more accessible commercial structure than a bespoke bilateral licensing negotiation that a smaller publisher would have little realistic chance of winning on its own. The news publisher guide covers this specific commercial mechanism, and its direct tie to citation rate, in more operational detail.
Is Blocking AI Crawlers Entirely a Viable Alternative Strategy?
For a narrow set of publishers with genuinely unique, high-value, hard-to-replicate content and a business model not primarily dependent on search or AI-referred traffic, certain premium subscription and paywalled research products fit this description, selectively blocking AI crawlers can be a coherent strategic choice, forcing any future access into a genuine bilateral licensing negotiation rather than surrendering it for free by default.
For the substantial majority of publishers, however, blocking simply means forfeiting AI citation and referral visibility with no corresponding licensing revenue materialising in exchange, precisely because of the search-training bundling problem described above, the blocking publisher loses the visibility without gaining any negotiating leverage that translates into an actual paid outcome. The realistic 2026 default for most publishers, articulated across every serious industry analysis, is allowing access while investing energy into maximising citation quality and owned-audience value from that access, rather than withholding it in the hope of a licensing negotiation that the current market structure makes unlikely to materialise.
How NotionCue Helps You Maximise Citation Value From the Access You Are Already Granting
Given that direct licensing revenue is realistically closed to most publishers under the current market structure, the value that remains fully within reach is maximising citation frequency, citation accuracy, and referral traffic from the AI crawler access most publishers are granting anyway, whether by explicit strategic choice or simply by not actively blocking it. That is a fundamentally different, and much more achievable, optimisation target than waiting for a licensing deal that the structural economics described throughout this article make unlikely for all but a small handful of brand-name publishers with genuine content scarcity.
The NotionCue llms.txt Generator builds a spec-compliant file from your actual content inventory, giving you direct, practical influence over which of your content AI systems prioritise when they do have access, the closest thing available to an opt-in signal that shapes outcomes, in the absence of the formal licensing leverage most publishers lack. The NotionCue Citation Tracker then measures whether that structuring effort is actually working, tracking your citation frequency and citation accuracy across all five major engines weekly, giving you the evidence needed to demonstrate genuine referral value internally, even in a market where a direct per-citation licensing payment from the AI companies themselves may never arrive.
Start your free NotionCue trial and build your llms.txt this week as the practical, achievable lever available to every publisher today, regardless of size or negotiating leverage, rather than waiting on a licensing market that current structural analysis suggests will not extend meaningfully beyond its existing brand-name participants.
If your organisation is approached directly by an AI company proposing a licensing arrangement, the Chicago Tribune and CNN litigation covered in 2026 press coverage offers a useful cautionary framework: get any claim about what content will and will not be used, and under what terms, in writing and specific, rather than accepting a general verbal assurance. Several of the active 2026 lawsuits centre on exactly this gap, where a company's informal assurance about limited or non-verbatim use diverged sharply from what the publisher's own subsequent investigation found had actually occurred.
Frequently Asked Questions About AI Content Licensing and Training Data Deals
Does adding a robots.txt disallow rule for GPTBot function as a legal opt-out from AI training?
Its legal force is genuinely unclear and actively contested, not a settled matter. Robots.txt is a technical, voluntary convention, not a codified legal instrument, and early court decisions in the EU specifically flagged robots.txt as a non-standardised, technically unsuitable mechanism for reliably exercising the formal rights-reservation options that frameworks like the EU's TDM exception provide. Treat a robots.txt disallow as a genuine technical access control, which most compliant crawlers do respect, rather than as an assured, litigation-proof legal opt-out from training use.
If I sign a licensing deal, does that guarantee my content will actually be cited more often in AI answers?
No, and this is a documented point of publisher frustration, not a hypothetical concern. Digiday's 2026 publisher scorecard reporting noted explicitly that partnership agreements do not guarantee citations, securing a formal licensing or revenue-share relationship with an AI company is a separate commercial and legal arrangement from that company's underlying citation-selection algorithm, which continues to evaluate content on the structural, freshness, and authority signals covered throughout this series regardless of whether a licensing relationship happens to exist in the background.
Should a small publisher pursue individual litigation against an AI company over unauthorised training use?
This is a significant legal and financial decision requiring qualified legal counsel specific to your jurisdiction and circumstances, and general content cannot responsibly substitute for that advice. What the current 2026 litigation landscape does show is a mixed picture: some cases have produced settlements that converted into ongoing commercial licensing relationships, as happened between Brazil's Folha and OpenAI, while others remain in active, extended discovery with no resolution yet in sight. The realistic operational takeaway for most smaller publishers is that this legal landscape remains genuinely unsettled, and building a resilient content and audience strategy that does not depend on a favourable outcome from litigation you may never pursue is the more reliable near-term path.