The model is forty years old and the application is new. Collecting societies have long licensed photocopying and digital reuse on behalf of publishers who could never chase each user individually, taking a fee from the user and distributing the proceeds back by share. Collective AI licensing extends that machinery to generative AI: one licence, one content repository, thousands of rightsholders, and an AI developer that pays once rather than negotiating separately with every publisher whose work it wants. For a publisher, the decision it creates is narrow and concrete. You opt in some or all of your catalogue, you choose which AI uses you are willing to license, and you wait to find out what the market pays.
What is collective AI licensing?
Collective AI licensing is a voluntary arrangement in which a collective management organisation licenses the content of many publishers to AI developers under a single agreement, collects a licence fee, and distributes the revenue back to participating rightsholders.
It exists because the direct licensing market has a shape problem. A few dozen brand-name publishers have negotiated bilateral deals with AI companies. Everybody else is offered nothing, and has no realistic route to a negotiation, because the transaction cost of a deal exceeds the value of any single small catalogue. Collective licensing solves that by aggregating: the thing being sold is not one publisher's archive but scale and clearance.
Sajeeda Merali, chief executive of the Professional Publishers Association, put the case plainly when the UK scheme launched. PPA members, she said, "represent trusted editorial brands from large international media companies to small, independent publishers, and many simply don't have the scale needed to influence the deal room. That's where collective licensing makes sense. It gives publishers strength in numbers and a practical way to make sure their content is used legally, transparently and is compensated properly."
How the UK collective AI licence works
The UK scheme is led by Publishers' Licensing Services, a non-profit collective management organisation owned and directed by four publishing trade bodies: the PPA, the Publishers Association, the Independent Publishers Guild and the Association of Learned and Professional Society Publishers. It has been built with the Copyright Licensing Agency and the Authors' Licensing and Collecting Society, and is branded the CLA Generative AI Solution.
There are two components. The first is a publisher and author backed licence that AI developers buy. The second is a content repository, built by the CLA, from which licensed developers can access the opted-in material. Press Gazette reported on 10 March 2026 that PLS had begun asking publishers to opt in, starting with book publishers at the London Book Fair, and that the CLA had "begun early discussions with AI companies".
The sequence matters for anyone reading the announcement as a revenue event. Stage one is publisher opt-in. Stage two, still ahead, is PLS and the CLA implementing the framework and enabling AI companies to license through the system. Opting in today puts your content in the shop window. It does not create a payment.
The scheme is also older than its launch date suggests. The CLA announced development of a Generative AI Training Licence in April 2025, slated for launch in the third quarter of that year, with Mat Pfleger, the CLA's chief executive, arguing that "training AI models on copyrighted content requires permission and compensation" and that collective licensing "can provide a market-based solution that is efficient and effective". The publisher opt-in stage opened roughly six months later than the original timetable. That slippage is not a scandal, but it is a useful calibration for how fast this route moves.
What the licence covers, and what it leaves out
PLS states that the licence is intended to cover the use of text for three things: training AI models, fine-tuning AI systems, and retrieval-augmented generation, where an AI tool draws on existing information to produce answers. It applies to both large and small language models.
Participation is voluntary at every level. Publishers decide whether to take part, which content to make available, and which category of generative AI use they are willing to license, with training and fine-tuning separable from retrieval-augmented generation. The scheme explicitly does not replace direct commercial agreements. PLS positions it as an additional route to market that runs alongside any deal a publisher has or wants to pursue.
Two limits are worth noting before you assume you are eligible. The scheme covers books, academic and scholarly publishing, magazines and specialist media, and professional and educational publishing, but only for publishers who license through the CLA rather than through NLA Media Access. NLA, which licenses content from more than 14,000 newspaper, magazine, newswire and digital-only titles, is developing its own separate AI proposition. If you are a news publisher licensing through NLA, this particular scheme is not your route, and you are waiting on a different one.
How it differs from direct deals, marketplaces and RSL
Four mechanisms are now competing to get publishers paid for AI use, and they are easy to confuse.
A direct deal is a bilateral contract between one publisher and one AI company. It pays the most and reaches the fewest.
A licensing marketplace, such as Microsoft's Publisher Content Marketplace or the various intermediary platforms, is a technology venue where publishers set terms and usage is metered per use. It is a distribution channel with a commercial operator behind it.
The RSL Collective applies the music-rights model to the web, expressing licensing terms in robots.txt and negotiating on publishers' behalf at scale. It is closest in spirit to collective licensing, but it is a new organisation building new infrastructure and new relationships.
A collecting society licence is the fourth, and its distinguishing feature is that the infrastructure already exists. PLS represents over 4,500 publishers and collected and distributed more than 48 million pounds in 2024/25 through its licensing and permissions schemes. ALCS represents over 125,000 writers and has paid out more than 700 million pounds since 1977. The rights registries, distribution mechanics and audit relationships are forty years old. That is the actual asset here, and Tom West, chief executive of PLS, made it the pitch: AI companies have complained that clearing permissions across many rightsholders is too difficult, and, as he put it, "we've got 40 years' experience of doing just that."
It is also worth separating collective licensing from collective standard setting. SPUR, the Standards for Publisher Usage Rights coalition formed in February 2026 by the Financial Times, The Guardian, Sky News, The Telegraph and the BBC, is not a collective licensing body and does not set prices for content use. PLS is in contact with SPUR and with RSL. These are complementary tracks, not rival ones.
What publishers might actually be paid
PLS has not published a rate. Its own answer is that it is too early to estimate the precise value of AI licensing and that it is developing pricing models based on established licensing benchmarks. The 48 million pounds figure it cites is historic distribution across all its schemes, not an AI forecast, and anyone quoting it as an AI number is misreading it.
What can be said is how the money would behave if it arrives. Collective licensing revenue is distributed by share, not by negotiation, which means it is proportional to the volume and value of what you contribute relative to everyone else. For a small publisher, that is simultaneously the point and the ceiling: you get access to a market you could not otherwise enter, and you get a small slice of it. West's framing to Press Gazette was that if the scheme works, "then there is an ongoing and sustainable revenue stream for publishers that simply wasn't available before". Ongoing and sustainable is the claim. Large is not.
The unresolved question is demand. West gave three reasons he expects AI companies to buy: access, because the trend towards a more closed web via tools like Cloudflare and TollBit is making unlicensed content harder to obtain; legal certainty, given the growing number of copyright cases, which he described as "a paradigm shift towards, actually, it's probably easier and quicker to licence"; and a growing recognition among AI companies that they cannot keep starving the people producing the content. Those are plausible arguments. They are not signatures. As of this writing no AI developer has been publicly named as a licensee of the UK scheme, and until one is, publishers are being asked to supply a market that has not yet bought anything.
The statutory licensing track running alongside it
The voluntary route has a compulsory cousin. Writing for Poynter on 9 March 2026, Anya Schiffrin and Roberta Carlini reported that policymakers in Europe, Brazil and other jurisdictions are exploring statutory licensing, which would require AI companies to pay publishers for journalism used to train their systems, past and future, rather than leaving it to private negotiation. The European Parliament was set to vote on 10 March 2026 on a proposal that could open the door to such a framework, and Brazil was weighing a draft bill expected in April.
The distinction is worth holding on to. Collective licensing is a market solution that depends on AI companies choosing to buy. Statutory licensing is a legal obligation that does not. One can be built now and may find no customers; the other would guarantee customers but does not exist yet, and may not.
The industry is not united on which to back. Danielle Coffey, president and chief executive of the News Media Alliance, told Poynter, "if we get the right verdicts, we will have a functional marketplace" - a position that puts litigation, not legislation, at the centre of the strategy. Meanwhile the pressure is arriving from several directions at once: Reuters reported in February 2026 that the European Publishers Council had filed an EU antitrust complaint over Google's AI Overviews. The United States has its own voluntary infrastructure rather than a statutory one, with Copyright Clearance Center having announced a voluntary, non-exclusive collective AI Systems Training License in March 2025 covering third-party content used to train AI systems.
For a publisher, the practical read is that none of these tracks is close to producing predictable revenue, and the sensible posture is to keep an option open on each rather than to bet the year on one.
What a collective licence does not solve
A collective licence prices your archive as an input. It does nothing about the moment a live AI agent fetches one of your pages to answer a question that has already been asked.
That distinction is not pedantry, and the licence's inclusion of retrieval-augmented generation makes it easy to blur. What the scheme licenses is access to opted-in content through the CLA's repository. What happens on your servers when a live search agent requests a URL in real time, renders it, extracts what it needs and returns an answer to a user who never arrives on your site is a separate event, on separate infrastructure, with separate economics. Your licence revenue does not scale with it, and your ad stack does not see it, because there is no browser to run the JavaScript that your monetisation depends on.
This is the layer blankspace works at, detecting live search agent requests at the CDN edge and monetising the response rather than the archive. It is not an alternative to a licensing conversation and should not be presented as one. A publisher can reasonably opt in to a collective licence for training and retrieval use of its back catalogue and separately monetise live agent traffic at the edge, because those are different transactions with different counterparties. The point is only that signing one does not settle the other, and a publisher who thinks the collective licence covers the whole AI relationship will be surprised.
Should you opt in?
For most publishers eligible for it, opting in is a low-cost option rather than a decision. The scheme is voluntary, non-exclusive, granular about which uses you permit, and explicitly designed to sit alongside direct deals. The main things to check before you do:
Read what you are granting, per use type. Training and fine-tuning are irreversible in a way retrieval is not. Content used to train a model cannot be withdrawn from that model later, whatever your opt-in status becomes.
Confirm your route. If you license through NLA Media Access rather than the CLA, this is not your scheme and you should ask NLA about its timetable.
Check the interaction with any direct deal. Non-exclusive in principle does not always mean unproblematic in practice if you have signed an exclusivity or most-favoured-nation clause elsewhere.
Do not model revenue from it. There is no published rate and no named licensee. Treat any income as upside, not as budget.
Keep your other levers. Blocking, access control, standards work through SPUR or RSL, and edge monetisation of live agent traffic all remain available and are not substitutes for one another.
Frequently asked questions
Who runs the UK collective AI licensing scheme?
Publishers' Licensing Services leads it, working with the Copyright Licensing Agency and the Authors' Licensing and Collecting Society. PLS is a non-profit owned and directed by the Professional Publishers Association, the Publishers Association, the Independent Publishers Guild and the Association of Learned and Professional Society Publishers. The licence itself is branded the CLA Generative AI Solution, and the content repository publishers opt into is built by the CLA.
What AI uses does the collective licence cover?
PLS states the licence is intended to cover the use of text for training AI models, fine-tuning AI systems, and retrieval-augmented generation, and that it applies to both large and small language models. Publishers choose which of those categories they are willing to license, so opting in for retrieval use does not oblige you to permit training.
Does opting in stop me doing a direct deal with an AI company?
No. The scheme is explicitly designed to operate alongside direct agreements and is described by PLS as an additional route to market rather than a replacement. The practical caveat is your own paperwork: if you have already signed an exclusivity or most-favoured-nation term with an AI company, check that against the collective licence before opting in.
How much will collective AI licensing pay publishers?
No rate has been published. PLS says it is too early to estimate the value and that it is developing pricing models based on established licensing benchmarks. The 48 million pounds PLS distributed in 2024/25 relates to its existing licensing and permissions schemes, not to AI, and should not be read as a forecast. Because collective revenue is distributed by share, a small publisher should expect a proportionally small slice of whatever the scheme earns.
Is collective licensing the same as statutory licensing?
No. Collective licensing is voluntary on both sides: publishers opt in, and AI developers choose whether to buy. Statutory licensing would be a legal requirement that AI companies pay for content used to train their systems, which policymakers in Europe, Brazil and elsewhere were reported to be exploring in 2026. The first exists and needs customers. The second would guarantee customers but does not yet exist.