Resource · Evidence

The evidence behind
A fair licensing collective

Each factual claim in the article, in one sentence, with the quote that supports it and its source: 212 claims.

Updated

Evidence

Claims, quotes and sources

This page backs the article A fair licensing collective. Quotes are copied from the source, and joined with an ellipsis where they come from two places in the same source. Non-English quotes are given in the original, with a translation. A claim that rests on something a source doesn't say, or on FORA's own status, has no quote. Two documents ILC shows publishers in its portal, the participation agreement and the Royalty Distribution Policy, have no public URL, so they are cited by section.

Primary source
The source says it: a scheme's own documents, a law, or FORA's own documentation and specification.
Reported by
Press or another third party reports it.
Our reading
An inference drawn from the sources, something a scheme's public pages don't say, or how a collective on FORA is meant to work. The reasoning is under the source.

Same money. Open boxes.

The collective

Primary source #journey-operator

A trade group, a publisher alliance or a company working for one runs the collective as its operator.

(operated by an Exchange operator, provider, or third party) handles all billing through its pluggable Billing Adapter

How money flows FORA documentation "The Billing Chain"

The quote describes the Exchange. FORA lets any organisation operate one; that a collective runs this way is the article's design.

What you put in

Our reading #journey-in-today

A pool today: the uses the scheme defines, every AI use or a fixed list.

The AI System uses the publisher’s licensed content under the ILC License, including for training, inference, and grounding, and tracks the usage required for reporting.

Publisher Onboarding Guide Internet Licensing Collective "How ILC licensing works"

A summary of the eight schemes, so its basis is the scorecard rows on this page, not one source. Row "Which uses do you license, and who picks them?" (#sc-uses-ilc to #sc-uses-ndp): ILC grants every AI use; ProRata, Kopinor and NDP fix the list; CLA lets the publisher pick between the two uses it lists.

Primary source #journey-in-fora

On FORA: your own terms, as simple or as detailed as you want.

A resource carries zero or more `LicenseTerm` entries, at most 32. Each term is a complete, self-contained access arrangement, and a resource with several terms is the normal case.

Licensing terms FORA documentation "Overview"

Picked

Our reading #journey-picked-today

A pool today: the AI company's or the operator's own system picks, and no scheme publishes how it picks for an answer.

The National Library shall determine which publications are used in the training of the various language models.

Agreement on the use of copyright-protected material in the training of Norwegian language models Kopinor and the National Library of Norway §6

A summary of the eight schemes, so its basis is the scorecard rows on this page, not one source. Row "Who picks whose content gets used" (#sc-picks-ilc to #sc-picks-ndp). The quote is the one contract that names the picker. GPT-NL publishes how TNO weights its training set, not how the model picks for an answer.

Primary source #journey-picked-fora

On FORA: the operator's search picks.

A Broker runs one catalog search, which its operator chooses per deployment. The search only proposes URIs: prices, license terms and signatures always come from the Exchange, so a search result cannot produce an offer the Exchange does not stand behind.

FORA documentation, the Broker: discovery providers FORA project no public URL

The operator can publish the method.

The query is configuration, not code.

FORA Server source, Broker search provider FORA project src/broker/internal/essearch/doc.go no public URL

The ranking is the operator's own search template, a configuration it can publish. FORA doesn't require publication. FORA Server isn't a public repository, so there is no URL.

Counted

A pool today: the payer's own report or engine counts, or no one counts at all.

The AI System sends the ILC Platform the licensing payment and a usage reporting file that includes its aggregated use of the publisher’s licensed content.

Publisher Onboarding Guide Internet Licensing Collective "How ILC licensing works"

A summary of the eight schemes, so its basis is the scorecard rows on this page, not one source. Row "Who counts the use your share rests on?" (#sc-counts-ilc to #sc-counts-ndp). CCC and CLA haven't said how they will count AI use.

Primary source #journey-counted-fora

On FORA: a signed record of every fetch, which the payer didn't write.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

The Exchange signs the offer; the AI company signs its acceptance (TransactionItem.agent_acceptance).

Split

Our reading #journey-split-today

A pool today: undisclosed rules or weights, in seven of the eight schemes; GPT-NL publishes its split rule.

Detailed calculations and technical implementation procedures are established in the Allocation Rules and are not publicly disclosed.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective §1 no public URL

A summary of the eight schemes, so its basis is the scorecard rows on this page, not one source. Row "Is the split rule published?" (#sc-split-ilc to #sc-split-ndp). GPT-NL: #sc-split-ndp.

Our reading #journey-split-fora

On FORA: a published formula and published weights.

FORA project statement FORA project no public URL

The article's design: the formula and the weights are the pool's published terms. The protocol leaves the money outside it, so the operator publishes them.

You see

Our reading #journey-see-today

A pool today: a statement, at most.

Publisher shall treat such summaries, royalty statements, and other nonpublic information made available through the Platform as confidential information of ILC.

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §7.5 no public URL

A summary of the eight schemes, so its basis is the scorecard rows on this page, not one source. ILC and ProRata send statements (#adds-statements); for Perplexity, CCC and Kopinor no statement was found.

Our reading #journey-see-fora

On FORA: your own records, and you can redo the sum.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

The protocol creates a signed record per fetch; showing a publisher its records and a statement with the totals is the operator's job in the article's design.

The main point

Our reading #journey-selection

None of the eight schemes publishes how an AI system picks whose content to use in an answer, and in a pool split by use that pick decides who gets paid.

The scorecard in the article FORA resources

Based on absence, across the scorecard row "Who picks whose content gets used" (#sc-picks-ilc to #sc-picks-ndp). GPT-NL publishes a training-set weighting, not an answer-time rule.

The collective

Our reading #journey-hero

A collective is fair when each member can see what the pool collects and check its own share.

FORA project statement FORA project no public URL

The article's premise. The parts that make it checkable on FORA are in the cards below; the schemes' answers are in the scorecard.

The formula

Our reading #journey-formula

Each AI company pays the greater of a share of its own revenue and a floor per fetch, up to a cap, so the pool never falls below the floor; the formula and the per-fetch floor are published, and each AI company's rate and cap are in its agreement with the operator.

A revenue share has no price at transaction time: its rate and its settlement are an agreement between the agent and the publisher outside this protocol.

Licensing terms FORA documentation "Revenue-share arrangements"

A design, not a source fact: min(cap, max(rate × its revenue, floor × fetches)); your pay = (pool − operator's cut) × your weight ÷ all members' weight. The protocol leaves rates and settlement to the parties' agreement.

Back to the topics

How collectives run today

Internet Licensing Collective (ILC)

Primary source #ilc-negotiated-fee

An AI company takes one blanket license for a negotiated fee, which ILC doesn't publish.

License fees are negotiated to reflect the economic value of the rights and content made available to AI licensees.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective §2.1 no public URL

The participation agreement: "ILC shall not be required to disclose exact negotiated rates, confidential pricing, or other commercially sensitive terms" (§2.2).

Primary source #ilc-reports-own-use

The AI company reports its own use.

The AI System sends the ILC Platform the licensing payment and a usage reporting file that includes its aggregated use of the publisher’s licensed content.

Publisher Onboarding Guide Internet Licensing Collective "How ILC licensing works"

Primary source #ilc-non-exclusive

Participation in ILC is non-exclusive.

Is participation exclusive? No. Participation is non-exclusive. You keep the right to license your content directly to anyone, and any existing licensing deals remain unchanged.

Frequently asked questions Internet Licensing Collective "Is participation exclusive?"

The participation agreement: designation "is non-exclusive and does not restrict direct licensing" (§3.5).

Primary source #ilc-exclude

A publisher can exclude a specific AI company.

Publisher may elect not to participate in any Collective License Agreement with a particular AI Company, or may withdraw some or all of its Designated Content from that AI Company’s Repertoire, at any time and for any reason, by authenticated notice through the Platform.

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §2.4 no public URL

Primary source #ilc-choose-uses

ILC's site tells publishers: "Choose which AI uses you authorize".

For what purposes Choose which AI uses you authorize.

For publishers Internet Licensing Collective "You decide"

Primary source #ilc-uses-agreement

Uses are all or nothing: the participation agreement sets the uses in each AI company's license.

Publisher agrees to participate in Collective License Agreements negotiated by ILC for the uses of Publisher’s Content expressly authorized under each applicable Collective License Agreement (the “Permitted Uses”).

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §2.1 no public URL

"All or nothing" is the article's summary: the publisher's lever in the agreement is to exclude an AI company (§2.4); no clause lets it keep an AI company and narrow the uses.

Primary source #ilc-uses-portal

The check in ILC's publisher portal accepts only a license declaration, the file where a publisher states its terms, that grants every AI use, training included.

The RSL declaration does not permit the required AI uses. Include a permits element with type="usage" and ai-all.

ILC publisher portal, website verification check Internet Licensing Collective no public URL

Shown to publishers in ILC's portal. ai-all is RSL's term for every AI use; ILC's public guide says the AI system uses the content "including for training, inference, and grounding".

Primary source #ilc-not-disclosed

The Royalty Distribution Policy says the detailed calculations are not publicly disclosed.

Detailed calculations and technical implementation procedures are established in the Allocation Rules and are not publicly disclosed.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective §1 no public URL

Repeated in §13.3.

ProRata / Gist

Primary source #prorata-fifty

ProRata's partner agreement puts 50% of the total revenue of its own answer products into one pool for all content partners.

“Revenue Share Percent” means the portion of Total Revenue (defined below) allocated for the pool of all content partners in the ProRata Offerings, which shall be fifty percent (50%) of the Total Revenue.

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

Primary source #prorata-attribution

ProRata's attribution engine splits the pool.

“Partner Per Answer Revenue Share” means the amount of revenue allocated to Partner for a given Answer from the Answer Engine based upon the proportional contribution from Partner’s Content to such Answer

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

The partner's share is "the product of Per Answer Revenue multiplied by the Revenue Share Percent and the Partner Per Answer Attribution Percent".

No outside AI company pays in.

“Total Revenue” means amounts earned and collected by ProRata from the ProRata Offerings including advertising, subscriptions, and sponsorships (if any)

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

The pool is a share of ProRata's own revenue from its own offerings; no third-party AI licensee is named on ProRata's or Gist's pages.

Perplexity Comet Plus

Primary source #pplx-announced

Perplexity announced Comet Plus in August 2025.

Comet Plus is a new subscription that gives Perplexity users access to premium content from a group of trusted publishers and journalists. Aug 25, 2025

Introducing Comet Plus Perplexity

Primary source #pplx-split

Subscription revenue goes to participating publishers, minus a small portion for Perplexity's compute costs, split by human visits, search citations and agent actions.

we’re distributing all of that revenue to participating publishers, minus a small portion for Perplexity’s compute costs. Comet Plus is the first compensation model that allocates revenue to our partners based on three types of internet traffic: human visits, search citations, and agent actions.

Introducing Comet Plus Perplexity

Our reading #pplx-only-payer

Perplexity runs the pool and is its only payer.

Comet Plus will be a $5 standalone subscription. Pro and Max subscribers get Comet Plus included in their memberships.

Introducing Comet Plus Perplexity

The pool is Perplexity's own subscription revenue, allocated by Perplexity; no other payer is described.

Primary source #pplx-partners

Its launch partners were Condé Nast, Fortune, Le Figaro, Le Monde, the Los Angeles Times and The Washington Post.

We are honored to launch Comet Plus with some of the world’s best-known, most respected sources of news and culture: Conde Nast, publishers of The New Yorker, Wired, Architectural Digest, Allure, Ars Technica, Bon Appétit, Condé Nast Traveler, Epicurious, GQ, Glamour, Pitchfork, Self, Teen Vogue, them, Vanity Fair, and Vogue. Fortune Le Figaro Le Monde The Los Angeles Times The Washington Post

Announcing Comet Plus launch partners Perplexity

The names after Condé Nast's list are a list of logos on the page.

Our reading #pplx-pricing-absent

Perplexity's pricing page doesn't list Comet Plus.

Pricing Perplexity

Absence: the pricing page mentions Comet and Comet Assistant, and Comet Plus nowhere.

Reported by #pplx-cnn-agreement

CNN's short agreement with Perplexity never became final.

“CNN’s short form agreement did not lead to a final agreement between the parties due to the inability to agree on multiple issues, including limits on Perplexity’s use of CNN content in its answers to users,” the lawsuit reads.

CNN Sues Perplexity for Unlawfully Distributing News Content TheWrap

CCC AI Systems Training License

Primary source #ccc-license

CCC describes a voluntary, non-exclusive collective license for training AI systems and for the trained system's later use as a service.

CCC’s AI Systems Training License is a voluntary, non-exclusive collective license for organizations training AI systems for uses external to their organization. It provides the rights to lawfully use a wide and diverse repertoire of content for training AI systems, as well as the subsequent use of the trained system as a service

AI Systems Training License: overview sheet Copyright Clearance Center

Our reading #ccc-nothing-named

CCC's public pages name no licensee, no price and no distribution rule for it.

Rightsholders receive royalty distributions electronically from CCC.

AI Systems Training License: overview sheet Copyright Clearance Center

Absence: this is all the overview sheet says about distribution, and no licensee or price appears on CCC's pages. A licensing tracker records "No public adoption figure found" (datalicenses.org).

CLA, with PLS and ALCS

Primary source #cla-building

The Copyright Licensing Agency is building a generative AI license with PLS, for publishers, and ALCS, for authors.

the initiative invites publishers to opt-in to a collective licence scheme, which has been developed along with the Copyright Licensing Agency (CLA) and the Authors' Licensing and Collecting Society (ALCS).

PLS offers new collective AI licensing opportunity to publishers Publishers' Licensing Services

Primary source #cla-regulated

CLA and PLS are regulated collectives.

PLS is regulated under the Collective Management of Copyright (EU Directive) Regulations 2016

Annual transparency report 2025–26 Publishers' Licensing Services "Compliance and code of conduct"

CLA: "CLA is a collective management organisation (CMO) and is controlled by its members" (CLA annual transparency report 2025, §6).

Primary source #cla-uses

Publishers decide the type of generative AI use they license, and PLS names two: training and retrieval.

Publishers decide: - whether to participate - which content they make available - the type of generative AI use they wish to license for (training and fine tuning, retrieval-augmented generation).

PLS offers new collective AI licensing opportunity to publishers Publishers' Licensing Services "Is the scheme mandatory?"

Primary source #cla-nla

Publishers that license through NLA Media Access can't join.

this licensing is available only to those publishers that license through CLA rather than NLA Media Access for Business and Public Sector licensing.

PLS offers new collective AI licensing opportunity to publishers Publishers' Licensing Services "Which publishers does this apply to?"

Our reading #cla-no-licensee

No AI licensee has been announced, and the license terms sit behind PLS's member login.

Publishers are now being invited to opt in to the scheme. Following this stage, PLS will work with the Copyright Licensing Agency and AI developers to implement the licensing framework and enable AI companies to license content through the system.

PLS offers new collective AI licensing opportunity to publishers Publishers' Licensing Services "What happens next?"

Absence: no licensee is named on CLA's or PLS's pages. The opt-in page in PLS's publisher area shows a sign-in screen only.

DPCMO

Primary source #dpcmo-regulated

Denmark's collective for press publishers is regulated.

Danske Pressepublikationers Kollektive Forvaltningsorganisation/Danish Press Publications Collective Management Organisation; forkortet DPCMO.

DPCMO (home page, in Danish) DPCMO

DPCMO is a collective management organisation under Danish law. Its approval to make extended-collective-licence agreements "gælder fra d.d. til og med den 26. september 2026" (applies until 26 September 2026), says the Ministry of Culture's decision; no renewal was found.

Primary source #dpcmo-vote

Its members vote in a general assembly.

Den øverste myndighed i DPCMO er generalforsamlingen. Medlemmerne er repræsenteret på generalforsamlingen ved en repræsentation pr. medievirksomhed.

Gennemsigtighedsrapport 2025 (transparency report, in Danish) DPCMO page 4

Translation: "The highest authority in DPCMO is the general assembly. The members are represented at the general assembly with one representation per media company."

Primary source #dpcmo-prorata-test

Its one signed AI agreement is a framework with ProRata, which its transparency report calls a test; each member opts in on its own.

DPCMO indgik i november 2024 letter of intent med ProRata.AI og i december 2025 en rammeaftale med henblik på at kunne lancere en dansk version af Giat.AI baseret udelukkende på licenseret indhold. Projektet er en test og vil kræve tilslutning fra en kritisk masse af danske presseudgivere.

Gennemsigtighedsrapport 2025 (transparency report, in Danish) DPCMO

Translation: "In November 2024 DPCMO entered a letter of intent with ProRata.AI, and in December 2025 a framework agreement, with a view to launching a Danish version of Giat.AI [sic, Gist.AI] based solely on licensed content. The project is a test and will require a critical mass of Danish press publishers to join." DPCMO's announcement: "Participation is free and non-exclusive". Its agreements with Mistral, Cloudflare and TollBit are letters of intent.

Primary source #dpcmo-openai

In February 2026, DPCMO took OpenAI to court.

OpenAI has trained its models on works from Danish press publications at least until the summer of 2024

A united Danish media industry takes OpenAI to court DPCMO

DPCMO's announcement is titled "A united Danish media industry takes OpenAI to court".

Kopinor and the National Library of Norway

Primary source #kopinor-scope

Kopinor licensed the National Library to train Norwegian language models on newspapers.

This agreement applies to the reproduction and use of text for the National Library’s training of language models from copyright-protected press publications published in Norway up to one year from the date of publication.

Agreement on the use of copyright-protected material in the training of Norwegian language models Kopinor and the National Library of Norway §2

Kopinor's announcement: "Avtalen regulerer bruk av innhold fra norske aviser i trening av norske språkmodeller." (The agreement governs the use of content from Norwegian newspapers in training Norwegian language models.)

Our reading #kopinor-only-price

It is the one published price among these schemes.

An annual total remuneration of NOK 45,000,000 (in 2026) shall be paid for the use of protected material under this agreement.

Agreement on the use of copyright-protected material in the training of Norwegian language models Kopinor and the National Library of Norway §17

Across the scorecard's "Who pays you today?" and "What does the operator keep?" rows, no other scheme publishes an amount. GPT-NL publishes its split rule and TNO's 50% share, but "TNO will determine the license fee for the Professional License (and any other paid license types) at its discretion" (template agreement §7.2).

Primary source #kopinor-paid

Kopinor paid NOK 44.1 million to MBL, the Norwegian media businesses' association.

44,1 millioner kroner ble etter avtalen utbetalt til Mediebedriftenes Landsforening (MBL) før sommeren.

Kopinors største utbetaling noensinne (Kopinor's largest payout ever, in Norwegian) Kopinor, via NTB

Translation: "NOK 44.1 million was paid to Mediebedriftenes Landsforening (MBL) under the agreement before the summer." MBL is the Norwegian media businesses' association.

Primary source #kopinor-closed

It has one licensee and covers Norwegian press only.

This agreement applies to the reproduction and use of text for the National Library’s training of language models from copyright-protected press publications published in Norway up to one year from the date of publication.

Agreement on the use of copyright-protected material in the training of Norwegian language models Kopinor and the National Library of Norway §2

The agreement is between Kopinor and the National Library alone, and covers press publications published in Norway.

NDP Nieuwsmedia and GPT-NL

Reported by #ndp-licensed

NDP Nieuwsmedia licensed articles from more than 30 news titles, plus the news agency ANP, to TNO to train GPT-NL.

Members of NDP Nieuwsmedia are providing access to their archives, which include articles from over 30 national and regional news titles. … News agency ANP is also participating

Dutch news publishers contribute to developing GPT-NL IO+

IO+: "Dutch news publishers are collaborating with the Netherlands Institute of Applied Scientific Research (TNO) to develop further GPT-NL".

Primary source #ndp-fifty

Contributors who choose revenue sharing split 50% of the net revenue from licensing the model; the template also offers no compensation, or a one-time payment. TNO keeps 50%.

Content Contributors can opt for one of three options: 1) no compensation is claimed, 2) a proportionate share of 50% of the Net Revenues (as defined below) generated with the GPT-NL Model and a proportionate discount on their professional license fee, or 3) a one-time upfront compensation and a proportionate discount on their professional license fee. … TNO will retain 50% of Net Revenues generated with the GPT-NL Model.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL

Primary source #ndp-reinvested

TNO's half will be re-invested in the model.

TNO’s share of the Net Revenues will be re-invested to cover cost of the further development and maintenance of the GPT-NL Model and the Research Facility

Content Contributor Agreement (template) TNO, GPT-NL

Primary source #ndp-rule

Each share is a token count times a quality multiplier.

The quantity of an individual dataset is determined by the number of Tokens that ends up in the Prepared Dataset. … The quality of a dataset depends on the oversampling multiplier used.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL §2.1.2–2.1.3

The share: "The share that a Content Contributor can expect is proportionate to the Data Value of the individual dataset as a relative percentage of the Data Value of the Prepared Dataset." (§2.1). Some of the PDF's formulas are images and didn't extract.

Primary source #ndp-choose-filter

Publishers choose what to contribute, and TNO filters and weights it.

Content Contributors each select the training content that they want to contribute to the GPT-NL project.

Annex 2: Training Content Protocol TNO, GPT-NL

TNO's weighting: "In the case of GPT-NL, high-quality Tokens can be repeated up to roughly 10 times for optimal performance", and where two sets overlap, "the data for which a Content Contributor Agreement was signed latest, will get deleted" (Annex 6 §2.1.2–2.1.3).

Primary source #ndp-exit

Leaving takes six months' notice, and models already trained may be used for two more years.

Either Party shall have the right to terminate this Agreement for convenience upon six (6) months prior written notice to the other Party with effect to the end of each contract year. … for a period of 2 years after the effective date of termination

Content Contributor Agreement (template) TNO, GPT-NL §13.2, §13.4

Our reading #ndp-template

The published terms are TNO's template; NDP's own contract isn't public.

iedere licentieovereenkomst met dataproviders gelden uniforme … regels en deze afspraken zijn openbaar te vinden op onze

GPT-NL progress report (in Dutch) TNO

Translation: "the same rules apply to every license agreement with data providers, and these arrangements can be found publicly on our [website]". The report's two-column layout splits the sentence. That NDP's contract follows the template is an inference from it; NDP's contract itself wasn't found.

Our reading #ndp-no-payout

No paying licensee of the model and no payout has been shown.

the participating news media receive no payment upfront and instead share in the proceeds if the model generates revenue.

GPT-NL respects copyright: cui bono? Open Future

Absence: no paying licensee of GPT-NL or payout to contributors was found in TNO's or NDP's public documents.

Our reading #ndp-closed

The license covers NDP's Dutch member titles only.

Members of NDP Nieuwsmedia are providing access to their archives, which include articles from over 30 national and regional news titles.

Dutch news publishers contribute to developing GPT-NL IO+

NDP licensed its members' archives; a publisher outside NDP isn't covered by it.

Music

Our reading #today-music

Music has run on pools for decades; a streaming service is a licensing collective with a player on top.

Every month, in every country, we calculate streamshare like this: your streams in that market ÷ total streams in that market = your share of that market’s royalty pool.

Loud & Clear: FAQ Spotify

The music comparison and its sources are on /resources/music-dsp/. Spotify's rule is one example of a pool split by a published formula.

Back to the topics

Around the world

Europe

Primary source #w-stim

Sweden: STIM runs a collective license for AI training and output; its first licensee is Songfox.

A new framework allows AI companies to train their systems on copyrighted music lawfully, with royalties flowing back to the original songwriters. Compensation flows both through model training and the downstream consumption of AI outputs. … The first company to operate under the new licence will be Songfox

STIM launches world's first collective AI music licence CISAC, for STIM

Primary source #w-gema

Germany: GEMA also publishes a licensing model at 30% of an AI provider's net income, with no licensee found.

The licensing approach provides a 30% share of all net income generated by the generative AI model or system of the provider.

Generative AI licensing model GEMA

No licensee of this model was found.

Our reading #w-kopiosto

Finland: Kopiosto's AI license covers a company's own staff, not AI developers.

Kopiosto’s copying licence will be extended to cover the use and prompting of artificial intelligence (AI) applications used internally by organisations.

Frequently asked questions about the AI extension Kopiosto

The FAQ defines an internal AI application as one "with access restricted only to the organisation’s staff". No Kopiosto license for AI developers was found.

Reported by #w-cedro

Spain: CEDRO signed as the exclusive Spanish partner of the RSL Collective, and has no license yet.

Nuestra Entidad se incorpora como socio exclusivo de la organización estadounidense en España

Un paso necesario para avanzar en el uso responsable de contenidos por la IA (in Spanish) ElDerecho.com

Translation: "Our organisation joins as the exclusive partner of the US organisation in Spain" (the organisation being the RSL Collective, named earlier in the article). No CEDRO AI license was found.

Primary source #w-cedro-name

The RSL Collective is a name ILC's partner page still uses.

LINC brings local publisher relationships and market expertise to RSL Collective

Partners Internet Licensing Collective

Reported by #w-dvp

France: DVP went to the competition authority in 2026 over Google's AI Overviews and AI Mode.

avoir saisi l’Autorité de la concurrence pour des manquements de Google à ses engagements. En cause, le déploiement en France des fonctionnalités d’intelligence artificielle « AI Overviews » et « AI Mode »

Droits voisins et IA : la presse attaque de nouveau Google devant l'Autorité de la concurrence (in French) Presse Agence

Translation: "[DVP announced it had] referred Google's breaches of its commitments to the competition authority. At issue: the rollout in France of the AI features AI Overviews and AI Mode."

Reported by #w-apig

France: APIG went to the competition authority in 2026 over AI Overviews and AI Mode.

L’organisme, qui représente près de 300 quotidiens français, demande à l’Autorité de la concurrence d’intervenir face au déploiement unilatéral des AI Overviews et de l’AI Mode.

Google AI Overviews : la presse française saisit l'Autorité de la concurrence (in French) Blog du Modérateur

Translation: "The body, which represents nearly 300 French dailies, asks the competition authority to act on the unilateral rollout of AI Overviews and AI Mode."

Primary source #w-epc

European Union: the European Publishers Council filed an antitrust complaint over Google's AI Overviews.

The European Publishers Council (EPC) has filed a formal complaint with the European Commission alleging that Google LLC and Alphabet Inc. are abusing their dominant position in general search services, in breach of Article 102 TFEU, through the deployment of AI Overviews and AI Mode within Google Search.

European Publishers Council files formal antitrust complaint against Google over AI Overviews and AI Mode European Publishers Council

Reported by #w-cjeu

An Advocate General at the EU's top court says chatbot answers can infringe press publishers' rights.

réutilisation en temps réel d'articles de presse par un chatbot d'intelligence artificielle peut porter atteinte aux droits des éditeurs lorsqu'elle franchit certaines lignes rouges, a conclu mardi l'avocat général à la Cour de Justice de l'Union européenne.

La réutilisation d'articles par un chatbot peut violer le droit de l'UE (in French) DH, via Belga

Translation: "real-time reuse of press articles by an AI chatbot can infringe publishers' rights when it crosses certain red lines, the Advocate General at the Court of Justice of the European Union concluded on Tuesday."

Asia

Reported by #w-kpf

Korea: the Korea Press Foundation sells news data for AI training from the outlets it manages in trust, 97 of them in 2024; each outlet sets its own price, and no buyer is named.

언론재단은 127개사의 뉴스저작물을 신탁 받아 관리 중인데 현재 신문, 인터넷신문, 방송사 97개 매체의 뉴스데이터를 이용할 수 있다고 밝혔다. … 구체적인 판매 요금은 각 언론사가 정한다.

언론재단, AI 학습용 뉴스데이터 판매 개시... 97개 매체 제공 (in Korean) Journalists Association of Korea

Translation: "The Press Foundation manages news works from 127 companies in trust, and said news data from 97 newspapers, internet newspapers and broadcasters is currently available. … Each news outlet sets the specific sale price." No buyer is named. The source is a 2024 report; no later one was found.

Reported by #w-daum

Korea: the portal Daum plans a pool of KRW 2 billion a year for partner outlets, split each quarter by how often its own AI summaries cite them.

구체적으론 향후 다음 검색에 더 적극 도입될 ‘AI 요약’에서 ‘AI의 인용 사용량’을 측정해 지급하는 방식이 제시됐다. 연 20억원 예산을 책정하고 ‘서비스 참여 의사’나 ‘데이터 활용 동의’를 한 매체에 대해 분기 단위로 기여도에 따라 금액을 분배한다는 구상이다.

'언론과 상생' 내세운 포털 다음, 'AI 인용'에 수익 지급키로 (in Korean) Journalists Association of Korea

Translation: "Specifically, it proposed measuring and paying by 'AI citation usage' in the 'AI summaries' to be introduced more widely in Daum search. The plan is to set a budget of KRW 2 billion a year and distribute it each quarter, by contribution, to outlets that agree to take part or to let their data be used."

Primary source #w-nsk

Japan: the newspaper association says AI training and AI search need permission.

日本新聞協会はこれまで、生成AIサービスが報道コンテンツを学習・利用する場合、報道機関の許諾を得るよう求めてきた。 … AI検索サービスでのコンテンツ利用についても権利者の許諾を得ることが原則であり

AI検索サービスに関する声明 (Statement on AI search services, in Japanese) Japan Newspaper Publishers & Editors Association

Translation: "The association has asked that generative AI services obtain news organisations' permission when they train on or use news content. … For the use of content in AI search services too, the principle is to obtain the rightsholder's permission." The article links the association's later statement, which says AI companies' transparency hasn't improved; the permission wording is in this earlier one.

Reported by #w-ani

India: the Delhi High Court refused ANI an interim ban against OpenAI, and ANI has appealed.

the Delhi High Court today pronounced its judgment on ANI’s application seeking interim injunction against OpenAI, the developer of AI chatbot ChatGPT. Justice Amit Bansal dismissed the application

Delhi High Court refuses interim injunction against OpenAI SpicyIP

The appeal: "The Delhi High Court has issued notice to OpenAI on an appeal by news agency Asian News International against a single judge’s refusal to grant it an interim injunction" (LawChakra).

Primary source #w-linc

LINC, a Singapore company, enrolls Asian publishers in ILC and other licensing channels; it names no AI licensee of its own.

LINC helps Asian publishers review readiness, enroll approved websites and subdomains, and manage participation in ILC. … © IntentBridges Pte. Ltd., Singapore.

Join the Internet Licensing Collective through LINC LINC (IntentBridges Pte. Ltd.)

The last part is an absence: LINC's pages name no AI licensee of its own.

Australia

Primary source #w-au-agency

Copyright Agency licenses business staff to use news in AI prompts; a license for AI developers is at member consultation.

In May 2025, we extended our Annual Business Licence – for businesses in sectors such as health sciences, finance, law and marketing – to allow staff to use licensed news media content in prompts for workplace AI tools. … Copyright Agency is now consulting with members on licensing their content for contextual datasets that improve outputs from AI tools or systems (e.g. via retrieval augmented generation or grounding), and developing AI models.

Collective licensing for AI Copyright Agency

The Americas

Reported by #w-ca-openai

Canada: news publishers are suing OpenAI, and the Ontario court kept jurisdiction.

The Ontario Superior Court of Justice (Commercial List) confirmed jurisdiction over OpenAI Inc. and related entities in this AI and technology dispute, allowing our clients' claims to proceed.

Canadian news media companies v OpenAI: jurisdiction motion Lenczner Slaght (counsel for the publishers)

Published by the publishers' counsel.

Reported by #w-br-meta

Brazil: ANJ and two other media associations won an injunction that bars Meta from using members' content for AI training.

A Justiça de São Paulo concedeu nesta terça-feira (3) uma liminar (decisão judicial provisória urgente) à Associação Nacional de Jornais (ANJ), à Associação Brasileira de Emissoras de Rádio e Televisão (Abert) e à Associação Nacional de Editores de Revistas (Aner) em ação contra a Meta. … inclusive para treinamento comercial de modelos de inteligência artificial.

Justiça proíbe Meta de usar conteúdo jornalístico sem autorização (in Portuguese) Jornal do Comércio

Translation: "The São Paulo court on Tuesday (3rd) granted an injunction (an urgent provisional ruling) to the National Newspaper Association (ANJ), the Brazilian Association of Radio and Television Broadcasters (Abert) and the National Association of Magazine Publishers (Aner) in a suit against Meta. … including for commercial training of artificial intelligence models." The injunction suspends Meta's new terms that would allow that use.

Reported by #w-br-bill

Brazil: a Senate-passed AI bill would require pay for training.

Cabe ao agente de IA que utilizar obras e conteúdos protegidos remunerar os respectivos titulares de direitos autorais e conexos pela utilização. - A remuneração pode ser negociada diretamente ou coletivamente.

IA e direitos autorais: o que diz o PL 2.338/23 aprovado pelo Senado (in Portuguese) Migalhas

Translation: "An AI agent that uses protected works and content must pay the respective copyright and related-rights holders for the use. Pay may be negotiated directly or collectively." The article's title calls the bill "aprovado pelo Senado" (passed by the Senate).

The closing statement

Our reading #w-transparency

Publishers' associations in Europe and Asia are asking AI platforms for transparency and equal treatment.

生成AI事業者による透明性の確保が進展していないため報道コンテンツの利用実態の把握が困難であり

AIの社会実装において、障害となる又は不十分な効果をもたらす規制・制度について (in Japanese) Japan Newspaper Publishers & Editors Association

Asia: the Japanese association says "because generative AI companies have made no progress on transparency, it is hard to know how news content is actually used" (quoted here). Europe: the Zurich declaration of the German-speaking publishers' associations asks platforms to stay neutral and favor no outlet (#w-zurich).

Reported by #w-zurich

The Zurich declaration of the German-speaking countries' publishers' associations says platforms may favor neither their own content nor that of individual outlets.

Internationale Online-Plattformen müssen die journalistische Pluralität achten und sich dieser gegenüber neutral verhalten, sie dürfen weder eigene Inhalte noch Inhalte einzelner Medien privilegieren

Verlegerverbände fordern Regeln für den Umgang von KI mit journalistischen Inhalten (in German) Netzwoche

Translation: "International online platforms must respect journalistic plurality and stay neutral towards it; they may favor neither their own content nor that of individual outlets." Netzwoche reports the declaration as the "Erklärung von Zürich" of "die Verlegerverbände aus dem deutschen Sprachraum".

Back to the topics

The scorecard

Who pays you today?

Primary source #sc-pays-prorata

ProRata / Gist: ProRata itself, from its own products.

“Total Revenue” means amounts earned and collected by ProRata from the ProRata Offerings including advertising, subscriptions, and sponsorships (if any)

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

Primary source #sc-pays-perplexity

Perplexity Comet Plus: Perplexity itself, from subscriptions.

Comet Plus will be a $5 standalone subscription. Pro and Max subscribers get Comet Plus included in their memberships.

Introducing Comet Plus Perplexity

Our reading #sc-pays-ccc

CCC: no licensee named.

No public adoption figure found.

CCC AI Licensing Suite datalicenses.org (tracker) "Adoption signals"

Absence: CCC's own pages name no licensee for the AI Systems Training License; a tracker reports the same.

Primary source #sc-pays-ndp

NDP / GPT-NL: TNO, from GPT-NL license revenue; no paying licensee shown.

Content Contributors can opt for one of three options: 1) no compensation is claimed, 2) a proportionate share of 50% of the Net Revenues (as defined below) generated with the GPT-NL Model … TNO will retain 50% of Net Revenues generated with the GPT-NL Model.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL

No paying licensee of the model is shown (#ndp-no-payout).

Our reading #sc-pays-fora

On FORA: no one; no collective runs on FORA yet.

FORA project statement FORA project no public URL

The FORA project's own status.

Which uses do you license, and who picks them?

Primary source #sc-uses-ilc

ILC: every AI use in each license; a publisher can exclude a licensee or a section.

Publisher may elect not to participate in any Collective License Agreement with a particular AI Company, or may withdraw some or all of its Designated Content from that AI Company’s Repertoire, at any time and for any reason, by authenticated notice through the Platform.

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §2.4 no public URL

Every AI use: the portal check (#ilc-uses-portal) and the agreement (#ilc-uses-agreement). A section: §2.4 lets the publisher "withdraw some or all of its Designated Content".

Primary source #sc-uses-prorata

ProRata / Gist: answers; no training; fixed.

providing Answers to End Users, on ProRata’s owned and operated AE Platform or through Distributed AI Search

ProRata Online Master Services Agreement (MSA) ProRata.ai §4

No training: "Publisher content supplied to ProRata for attribution is never used for model training" (ProRata, for publishers).

Primary source #sc-uses-perplexity

Perplexity Comet Plus: pays on visits, citations and agent actions; the publisher's choice of uses isn't stated.

allocates revenue to our partners based on three types of internet traffic: human visits, search citations, and agent actions.

Introducing Comet Plus Perplexity

Not stated: the scheme's public pages don't say.

Primary source #sc-uses-cla

CLA / PLS: the publisher picks training or retrieval.

Publishers decide: - whether to participate - which content they make available - the type of generative AI use they wish to license for (training and fine tuning, retrieval-augmented generation).

PLS offers new collective AI licensing opportunity to publishers Publishers' Licensing Services "Is the scheme mandatory?"

Primary source #sc-uses-ndp

NDP / GPT-NL: training only; fixed.

license content to TNO for the training and future updating of the GPT-NL Model.

Annex 2: Training Content Protocol TNO, GPT-NL

Licensees of the model may not use it for "Media Integrator Services" (Annex 7).

Primary source #sc-uses-fora

On FORA: the publisher picks, per term; a pool leaves training out.

A resource carries zero or more `LicenseTerm` entries, at most 32. Each term is a complete, self-contained access arrangement, and a resource with several terms is the normal case.

Licensing terms FORA documentation "Overview"

Each term carries its own use restriction. That a pool term leaves training out is the article's design: training on a held copy leaves no fetch record to check.

Who picks whose content gets used, and is the method published?

Our reading #sc-picks-ilc

ILC: the AI company; method not published.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective no public URL

Absence: ILC's public pages, agreement and policy were searched for "rank", "relevance", "select", "neutral", "bias", "prefer" and "steer"; none describes how a licensee picks sources (#selection-ilc-silent).

Our reading #sc-picks-prorata

ProRata / Gist: ProRata's own engine; method not published.

Answers come from your content first, with 1,600+ licensed titles behind it.

Gist Answers ProRata.ai

ProRata's engine picks; no selection method is published. ProRata's worked attribution example says "this is not a universal scoring formula".

Primary source #sc-picks-ccc

CCC: the AI company, from content it already buys; method not stated.

Organizations purchase and rely on CCC’s AI Systems Training License for rights to reuse the content they purchase or subscribe to for training AI systems.

AI Systems Training License: overview sheet Copyright Clearance Center

Not stated: the scheme's public pages don't say.

Primary source #sc-picks-cla

CLA / PLS: the publisher chooses what to offer; the AI company's pick is not stated.

Publishers decide: - whether to participate - which content they make available - the type of generative AI use they wish to license for (training and fine tuning, retrieval-augmented generation).

PLS offers new collective AI licensing opportunity to publishers Publishers' Licensing Services "Is the scheme mandatory?"

"which content they make available" is the publisher's choice. Not stated: the scheme's public pages don't say.

Our reading #sc-picks-dpcmo

DPCMO: ProRata's engine, in the test; DPCMO's documents don't say.

DPCMO indgik i november 2024 letter of intent med ProRata.AI og i december 2025 en rammeaftale med henblik på at kunne lancere en dansk version af Giat.AI baseret udelukkende på licenseret indhold. Projektet er en test og vil kræve tilslutning fra en kritisk masse af danske presseudgivere.

Gennemsigtighedsrapport 2025 (transparency report, in Danish) DPCMO

Translation: "In November 2024 DPCMO entered a letter of intent with ProRata.AI, and in December 2025 a framework agreement, with a view to launching a Danish version of Giat.AI [sic, Gist.AI] based solely on licensed content. The project is a test and will require a critical mass of Danish press publishers to join." The test runs on ProRata's Gist; DPCMO's documents say nothing about selection.

Primary source #sc-picks-ndp

NDP / GPT-NL: you choose what to contribute; TNO filters and weights it for training, by a published rule.

Content Contributors each select the training content that they want to contribute to the GPT-NL project.

Annex 2: Training Content Protocol TNO, GPT-NL

The weighting rule: "In the case of GPT-NL, high-quality Tokens can be repeated up to roughly 10 times for optimal performance" (Annex 6 §2.1.3). It governs the training set, not what the model draws on in an answer.

Primary source #sc-picks-fora

On FORA: the operator's search; the ranking rule is the operator's own, and it can publish it.

A Broker runs one catalog search, which its operator chooses per deployment. The search only proposes URIs: prices, license terms and signatures always come from the Exchange, so a search result cannot produce an offer the Exchange does not stand behind.

FORA documentation, the Broker: discovery providers FORA project no public URL

The template that ranks results is the operator's configuration (#selection-template). FORA doesn't require publication.

Who counts the use your share rests on?

Primary source #sc-counts-ilc

ILC: the AI company, in its own report.

The AI System sends the ILC Platform the licensing payment and a usage reporting file that includes its aggregated use of the publisher’s licensed content.

Publisher Onboarding Guide Internet Licensing Collective "How ILC licensing works"

Primary source #sc-counts-prorata

ProRata / Gist: ProRata's attribution engine.

“Partner Per Answer Revenue Share” means the amount of revenue allocated to Partner for a given Answer from the Answer Engine based upon the proportional contribution from Partner’s Content to such Answer

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

Our reading #sc-counts-perplexity

Perplexity Comet Plus: Perplexity.

allocates revenue to our partners based on three types of internet traffic: human visits, search citations, and agent actions.

Introducing Comet Plus Perplexity

Perplexity allocates by traffic it observes in its own products; no outside count is described.

Primary source #sc-counts-ccc

CCC: not stated for AI; licensee surveys in CCC's other licenses.

These allocations are based on licensee surveys and other data collection activities, which are agreed to by the licensees as a condition of the license.

Comment to the US Copyright Office on mass digitization Copyright Clearance Center

An older CCC filing, about its annual licenses.

Primary source #sc-counts-cla

CLA / PLS: not stated for AI; licensee reports, audits and surveys in CLA's other licenses.

The allocation methodology of CLA is based on a combination of: usage reports by licensees, usage audits, surveys, and other methods as determined from time to time by the CLA Board

Distribution Charter Publishers' Licensing Services §4.1.1

Primary source #sc-counts-ndp

NDP / GPT-NL: TNO, once: your tokens in the training set, and their quality.

The quantity of an individual dataset is determined by the number of Tokens that ends up in the Prepared Dataset. … The quality of a dataset depends on the oversampling multiplier used.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL §2.1.2–2.1.3

"Every new GPT-NL Model with a Professional License will have its own revenue sharing calculation." Use in answers isn't measured.

Primary source #sc-counts-fora

On FORA: signed fetch records, with AI companies' reports checked against them.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

The AI company also signs its acceptance (TransactionItem.agent_acceptance). Checking reports against the records is the operator's job in the article's design.

Is the split rule published?

Primary source #sc-split-ilc

ILC: principles; the calculations "are not publicly disclosed".

Detailed calculations and technical implementation procedures are established in the Allocation Rules and are not publicly disclosed.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective §1 no public URL

Primary source #sc-split-prorata

ProRata / Gist: the formula, not the attribution weights.

the product of Per Answer Revenue multiplied by the Revenue Share Percent and the Partner Per Answer Attribution Percent. - “Per Answer Revenue” means the Total Revenue divided by the number of Total Valid Answers in a given period.

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

On the weights, ProRata's own worked example says "this is not a universal scoring formula" (How it works).

Primary source #sc-split-perplexity

Perplexity Comet Plus: three categories of traffic.

allocates revenue to our partners based on three types of internet traffic: human visits, search citations, and agent actions.

Introducing Comet Plus Perplexity

Payouts aren't disclosed: "we are not disclosing any of those numbers" (Digiday interview with Perplexity's head of publisher partnerships).

Primary source #sc-split-cla

CLA / PLS: general splits; no AI rule yet.

Journals 76.00 23.00 1.00

Distribution Charter Publishers' Licensing Services revenue shares table

One row of the published table of shares for standard licenses. No AI-specific rule is published.

Our reading #sc-split-dpcmo

DPCMO: not online.

vedtog foreningen på generalforsamlingen i 2024 en midlertidig fordelingspolitik for 2024- 2025. Fordelingen afventer afklaring med autorerne.

Gennemsigtighedsrapport 2025 (transparency report, in Danish) DPCMO

Translation: "the association adopted a temporary distribution policy for 2024–2025 at its 2024 general assembly. Distribution awaits clarification with the authors." The policy itself was not found online.

Primary source #sc-split-kopinor

Kopinor: general guidelines; the split among newspapers not found.

Vederlag til rettighetshaverne fordeles etter retningslinjer som er vedtatt av Kopinors representantskap.

Årsmelding 2025 (annual report, in Norwegian) Kopinor §9.1

Translation: "Remuneration to rightsholders is distributed under guidelines adopted by Kopinor's council of representatives." How MBL splits the AI money among newspapers was not found.

Primary source #sc-split-ndp

NDP / GPT-NL: yes, in TNO's template: token count times a quality multiplier.

The share that a Content Contributor can expect is proportionate to the Data Value of the individual dataset as a relative percentage of the Data Value of the Prepared Dataset.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL §2.1

The two factors: "The quantity of an individual dataset is determined by the number of Tokens that ends up in the Prepared Dataset." and "The quality of a dataset depends on the oversampling multiplier used." (§2.1.2–2.1.3).

Our reading #sc-split-fora

On FORA: every weight and threshold is published.

FORA project statement FORA project no public URL

The article's design: the allocation rule is part of the pool's published terms; the protocol carries none.

What does the operator keep?

Primary source #sc-keeps-ilc

ILC: its costs, "projected at or below 10%".

After operating expenses, projected at or below 10% and consistent with comparable collective licensing organizations, the remaining revenue passes directly to rights holders.

The Internet Licensing Collective (ILC) Launches with Yahoo Scout as First Commercial Partner (press release) Internet Licensing Collective

A projection: the agreement's costs include "reasonable reserves for anticipated operating expenses and obligations" (§1.1(a)), with no ceiling.

Primary source #sc-keeps-prorata

ProRata / Gist: 50%.

“Revenue Share Percent” means the portion of Total Revenue (defined below) allocated for the pool of all content partners in the ProRata Offerings, which shall be fifty percent (50%) of the Total Revenue.

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

Primary source #sc-keeps-cla

CLA / PLS: 11% (CLA) and 6% (PLS) on standard licenses; not stated for AI.

CLA deducts an administration fee, prior to distribution to the CMOs as set out in paragraph 2.2 above, of: 3.1.1 11% in the case of revenue collected from all licences … an administration fee of 6%.

Distribution Charter Publishers' Licensing Services §3.1, §3.3

The charter predates the AI license; whether these fees apply to AI revenue is not stated.

Primary source #sc-keeps-ndp

NDP / GPT-NL: 50%, re-invested in the model.

TNO will retain 50% of Net Revenues generated with the GPT-NL Model.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL

The template: "TNO’s share of the Net Revenues will be re-invested to cover cost of the further development and maintenance of the GPT-NL Model and the Research Facility" (#ndp-reinvested).

Our reading #sc-keeps-fora

On FORA: the operator sets its cut and publishes it.

A revenue share has no price at transaction time: its rate and its settlement are an agreement between the agent and the publisher outside this protocol.

Licensing terms FORA documentation "Revenue-share arrangements"

The article's design: the cut is part of the pool's published terms.

Is it regulated, and do members vote?

Primary source #sc-governance-ilc

ILC: not regulated; no member vote; a publisher steering committee.

Participation under this Agreement does not, by itself, make Publisher a member of ILC or confer any voting, governance, or other membership rights in ILC

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §2.1 no public URL

The committee: "The ILC Publisher Steering Committee oversees the licensing framework, safeguards, and economics of the collective" (FAQ); the policy calls its role "advisory" (§13.1). Not regulated: ILC is "a Tennessee nonprofit corporation", and no US statute regulates a voluntary collective (a reading).

Perplexity Comet Plus: a private company; no vote found.

Through this strategic content licensing agreement, premium journalism from USA TODAY and the USA TODAY Network's 200+ local publications will be integrated into Perplexity's AI-powered search experiences.

Welcoming Gannett to the Perplexity Publisher Program Perplexity

Perplexity's publisher deals are bilateral agreements; no governance role for publishers was found.

Primary source #sc-governance-cla

CLA / PLS: regulated; publishers vote through trade associations.

PLS is regulated under the Collective Management of Copyright (EU Directive) Regulations 2016

Annual transparency report 2025–26 Publishers' Licensing Services "Compliance and code of conduct"

PLS's governance page: "Owned and directed by: Association of Learned and Professional Society Publishers" and other trade associations.

Primary source #sc-governance-dpcmo

DPCMO: regulated; members vote.

Den øverste myndighed i DPCMO er generalforsamlingen. Medlemmerne er repræsenteret på generalforsamlingen ved en repræsentation pr. medievirksomhed.

Gennemsigtighedsrapport 2025 (transparency report, in Danish) DPCMO page 4

Translation: "The highest authority in DPCMO is the general assembly. The members are represented at the general assembly with one representation per media company." Regulated: #dpcmo-regulated.

Primary source #sc-governance-kopinor

Kopinor: regulated; member organizations vote.

Kopinor er omfattet av lov om kollektiv forvaltning av opphavsrett mv. og står under tilsyn av Patentstyret.

Årsmelding 2025 (annual report, in Norwegian) Kopinor

Translation: "Kopinor is covered by the act on collective management of copyright and is supervised by the Norwegian Industrial Property Office." Its distribution guidelines are adopted by its council of representatives (§9.1).

Primary source #sc-governance-ndp

NDP / GPT-NL: not regulated; no direct vote; a contributors' Content Board.

that may have a substantial negative impact on the protection provided to Content Contributors under the Content Contributor Agreement, may only be taken with the agreement of the member on behalf of the Content Board, which will not be unreasonably withheld.

Annex 3: Governance Charter TNO, GPT-NL §2.5.3.4

TNO decides; the Content Board's member must agree to changes that weaken contributor protections. NDP is a trade association and TNO a public research body; neither is a regulated collective (a reading).

Our reading #sc-governance-fora

On FORA: not regulated; each operator sets its governance.

FORA project statement FORA project no public URL

FORA is a protocol; it defines no governance for a pool.

Does it reach copies an AI company took before any license?

Primary source #sc-held-ilc

ILC: yes.

Under the ILC License, these terms and restrictions apply wherever an authorized AI System obtains or uses the content, including content obtained directly from the publisher or through third parties, acquired before the RSL declaration was published, or otherwise held without a license governing its AI use.

Publisher Onboarding Guide Internet Licensing Collective "What an RSL declaration does"

ILC's enforcement mandate starts at designation: "occurring on or after the date such Content became Designated Content" (agreement §5.4(g)).

Primary source #sc-held-cla

CLA / PLS: no; "No retrospective cover".

No retrospective cover

PLS opt-in licences: flyer Publishers' Licensing Services "GenAI Product Licensing"

The flyer's layout makes the column assignment slightly ambiguous.

Primary source #sc-held-kopinor

Kopinor: the library's own collections, yes; earlier training of its base models, no.

har Nasjonalbiblioteket og Kopinor blitt enige om vilkårene for å benytte Nasjonalbibliotekets samlinger av beskyttede publikasjoner til slik maskinlæring.

Trening av norske språkmodeller (in Norwegian) Kopinor

Translation: "the National Library and Kopinor have agreed the terms for using the National Library's collections of protected publications for such machine learning." The agreement: "It does not affect any pre-training that the base models used by the National Library may have undergone." (§11).

Our reading #sc-held-ndp

NDP / GPT-NL: no; it covers GPT-NL's own training only.

license content to TNO for the training and future updating of the GPT-NL Model.

Annex 2: Training Content Protocol TNO, GPT-NL

The license is to TNO for training GPT-NL; it says nothing about other developers' earlier copies. Open Future: "GPT-NL trains only on data it has positive authorization to use".

Our reading #sc-held-fora

On FORA: no; a copy already held leaves no record to check.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

FORA records each fetch it brokers. A model trained on a copy it already holds makes no fetch.

Under the table

Primary source #sc-non-exclusive

Six of the eight say participation is non-exclusive.

Is participation exclusive? No. Participation is non-exclusive. You keep the right to license your content directly to anyone, and any existing licensing deals remain unchanged.

Frequently asked questions Internet Licensing Collective "Is participation exclusive?"

ILC (quoted). ProRata: "No. It is non-exclusive and fits alongside anything else in your AI strategy" (Gist Answers). CCC: "a voluntary, non-exclusive collective license". PLS: "the non-exclusive right for PLS" (terms and conditions §3.2). DPCMO: "Participation is free and non-exclusive". GPT-NL: "The Content Contributor may license the Contributor Training Content to third parties at its discretion" (template agreement). Perplexity and Kopinor don't say.

Back to the topics

What FORA adds

Our reading #adds-record-schemes

In the six schemes that say how use is counted, the AI company reports its own use, the operator or the licensee counts it, or no one counts per use; CCC and CLA haven't said for AI.

The AI System sends the ILC Platform the licensing payment and a usage reporting file that includes its aggregated use of the publisher’s licensed content.

Publisher Onboarding Guide Internet Licensing Collective "How ILC licensing works"

Basis: the scorecard row "Who counts the use your share rests on?" (#sc-counts-ilc to #sc-counts-ndp). ILC: the AI company. ProRata and DPCMO: ProRata's engine. Perplexity: its own traffic. NDP: TNO, the licensee. Kopinor: a lump sum.

Primary source #adds-record-fora

On FORA, every fetch is a transaction for one URL; the operator's Exchange signs the offer, and the AI company signs its acceptance.

The agent's detached acceptance signature over this item's `offer`.

Protocol reference: FORA v1 FORA documentation TransactionItem.agent_acceptance

The Exchange's signature covers the entire offer (Offer.signature, #journey-counted-fora).

Our reading #adds-audit-only

The schemes' public terms offer at most an audit or a dispute right.

Partner may request to have an independent third party confidentially inspect the Records if it reasonably disputes Revenue Share paid under this Agreement.

ProRata Online Master Services Agreement (MSA) ProRata.ai §6

ProRata allows an inspection in a dispute; ILC says it "supports auditing of license compliance" (publishers page); CLA's general method includes "usage audits". No scheme describes checking each report against a record it didn't get from the payer.

Our reading #adds-reconcile

On FORA, the operator reconciles each AI company's report with the fetch records before it splits the pool.

The agent reports what it consumed, not what it estimated: an Exchange MUST NOT refuse a report because this differs from the estimate, in either direction.

Protocol reference: FORA v1 FORA documentation UsageReport.consumed_quantity

The protocol binds each report to its transaction; the reconciliation is the operator's job in the article's design.

Our reading #adds-recompute-seven

Seven of the eight don't publish enough to redo the sum.

Detailed calculations and technical implementation procedures are established in the Allocation Rules and are not publicly disclosed.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective §1 no public URL

Basis: the scorecard row "Is the split rule published?" (#sc-split-ilc to #sc-split-ndp).

Our reading #adds-recompute-gptnl

GPT-NL publishes its rule; its public documents don't say whether a contributor gets the token counts and quality scores the sum needs.

The share that a Content Contributor can expect is proportionate to the Data Value of the individual dataset as a relative percentage of the Data Value of the Prepared Dataset.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL §2.1

The rule is published (#sc-split-ndp). Absence: no contributor statement or report of token counts and scores was found; the agreement itself is confidential (template §11).

Our reading #adds-recompute-fora

On FORA, your statement gives your weight, the total weight, the pool, the operator's cut and the version of the rule.

FORA project statement FORA project no public URL

The article's design for the operator's statement.

Primary source #adds-statements

ProRata sends monthly statements down to the article; ILC sends periodic ones and makes them confidential; none promises data per fetch.

ProRata will deliver to Partner (including by making available to partner a web-based access to the information) a statement of Partner Revenue Share for the preceding month

ProRata Online Master Services Agreement (MSA) ProRata.ai §6, "Reporting"

ProRata's statement goes to "Partner-level", publication and article detail. ILC: "ILC shall provide Publisher with periodic royalty statements through the Platform" (§4.1), and statements are "confidential information of ILC" (§7.5). "None promises data per fetch" is an absence across the eight.

Our reading #adds-own-data

On FORA, you see each of your transactions, and the report filed against it, as it arrives.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

The protocol creates a record per transaction; showing it to the publisher as it arrives is the operator's dashboard in the article's design.

Primary source #adds-own-terms

Your own priced terms beside the pool's: join the pool with one term for everything, or a term per section and per use; terms you price yourself run next to it, and the same fetch records back both.

A resource carries zero or more `LicenseTerm` entries, at most 32. Each term is a complete, self-contained access arrangement, and a resource with several terms is the normal case.

Licensing terms FORA documentation "Overview"

Our reading #adds-limit

A record proves the fetch; what a model did with the page afterward comes from the AI company's report, and an audit checks it.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

The signatures cover what was offered and accepted for one fetch. Use inside the model leaves no record outside the AI company.

Back to the topics

Who decides which publisher gets used

Our reading #selection-fee-same

An AI company pays the same fee whichever publisher its system picks, so the split between publishers follows the pick.

AI companies track when your content is used, whether for grounding, citation, or other uses that help shape the output.

For publishers Internet Licensing Collective

A share measured from the licensee's use rises and falls with the licensee's choices; the fee is set by the license.

Primary source #selection-prorata-engine

ProRata's own engine chooses the sources.

Answers come from your content first, with 1,600+ licensed titles behind it.

Gist Answers ProRata.ai

Our reading #selection-ilc-silent

ILC's public pages, its agreement and its policy don't say how a licensee chooses.

ILC Royalty Distribution Policy, shown in ILC's publisher portal Internet Licensing Collective no public URL

Based on absence: ILC's public pages, the agreement and the policy were searched for "rank", "relevance", "select", "neutral", "bias", "prefer" and "steer". The only hit near the topic is the policy's principle "Apply Consistent, Content-Neutral Standards" (§3), which governs ILC's own allocation, not the AI system's choice of sources.

Primary source #selection-gptnl

GPT-NL comes closest, for training only: it publishes how TNO filters and weights each contributor's text.

In the case of GPT-NL, high-quality Tokens can be repeated up to roughly 10 times for optimal performance.

Annex 6: Revenue Sharing Mechanism TNO, GPT-NL §2.1.3

On overlap: "the data for which a Content Contributor Agreement was signed latest, will get deleted" (§2.1.2). This weights the training set; it isn't an answer-time rule.

Primary source #selection-template

The operator writes the ranking rule, so the collective holds it and can publish it.

The query is configuration, not code. … The template, stored in the cluster (Config.TemplateID) or sent with every search (Config.TemplateSource), decides which fields are searched, how they rank and how each filter applies.

FORA Server source, Broker search provider FORA project src/broker/internal/essearch/doc.go no public URL

FORA Server's Elasticsearch catalog search. FORA Server isn't a public repository, so there is no URL. FORA doesn't require the operator to publish the template.

Primary source #selection-gutefrage

Gutefrage, which runs FORA on helpster.de, uses the Elasticsearch search engine.

Gutefrage, as an implementor, chooses to have Elasticsearch

Implementer information FORA project no public URL

Implementer information given to the FORA project; no public URL.

Primary source #selection-per-url

The search only proposes URLs: each fetch is licensed on its publisher's terms.

Elasticsearch supplies candidate URLs and nothing else. Every URL then takes the Broker's normal per-URL route (publisher manifest, Exchange discovery, offer verification), so a stale hit cannot produce an offer the Exchange does not stand behind.

FORA Server source, Broker search provider FORA project src/broker/internal/essearch/doc.go no public URL

A catalog entry names one resource by its domain and exact path (ResourceEntry, /reference/proto-fora/#resourceentry).

Primary source #selection-recorded

Each fetch is recorded as its own transaction for one resource.

each item is atomic, and its integrity is per resource. Each offer carries the signature of the Exchange that issued it, covering price, terms, expiry and `exchange`; each item carries the agent's acceptance of that one offer

Protocol reference: FORA v1 FORA documentation TransactionItem

The AI company's model still decides what to cite.

FORA project statement FORA project no public URL

The search proposes URLs and the Exchange signs offers for them; what the AI system then grounds on and cites happens inside its model, which the operator doesn't control.

Back to the topics

Where others are ahead of FORA

They run, and some pay

Primary source #ahead-prorata-shares

ProRata shares revenue with its partners under a public agreement.

“Revenue Share Percent” means the portion of Total Revenue (defined below) allocated for the pool of all content partners in the ProRata Offerings, which shall be fifty percent (50%) of the Total Revenue.

ProRata Online Master Services Agreement (MSA) ProRata.ai definitions

Our reading #ahead-none-on-fora

No collective runs on FORA yet.

FORA project statement FORA project no public URL

The FORA project's own status.

A legal mandate, and enforcement

Primary source #ahead-regulated

CLA, PLS, DPCMO and Kopinor are regulated: the law sets what they must disclose, and a supervisor oversees them.

PLS is regulated under the Collective Management of Copyright (EU Directive) Regulations 2016

Annual transparency report 2025–26 Publishers' Licensing Services "Compliance and code of conduct"

DPCMO: #dpcmo-regulated. Kopinor: "Kopinor er omfattet av lov om kollektiv forvaltning av opphavsrett mv. og står under tilsyn av Patentstyret." (Kopinor is covered by the act on collective management of copyright and is supervised by the Norwegian Industrial Property Office; Kopinor annual report.)

Primary source #ahead-ilc-attorney

ILC's participation agreement lets it sue unlicensed AI companies in a publisher's name.

to bring such proceedings in the name of Publisher, in the name of ILC, or in the name of such other person, as applicable law permits

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §5.4(c) no public URL

The same clause appoints ILC "as Publisher’s true and lawful attorney-in-fact".

FORA is a protocol: it has no mandate, and it takes no one to court.

FORA project statement FORA project no public URL

FORA defines a wire format; enforcement is left to the parties.

Copies AI companies already hold

Primary source #ahead-ilc-held

ILC's license reaches content that AI companies already hold, "acquired before the RSL declaration was published".

Under the ILC License, these terms and restrictions apply wherever an authorized AI System obtains or uses the content, including content obtained directly from the publisher or through third parties, acquired before the RSL declaration was published, or otherwise held without a license governing its AI use.

Publisher Onboarding Guide Internet Licensing Collective "What an RSL declaration does"

Primary source #ahead-rsl

RSL is the format ILC uses for that declaration.

The ILC licensing platform uses the open RSL (Really Simple Licensing) standard to enable publishers specify which content is available for AI licensing.

Frequently asked questions Internet Licensing Collective "How is the ILC related to RSL?"

Primary source #ahead-training-licensed

CCC, CLA, Kopinor and NDP license training.

CCC’s AI Systems Training License is a voluntary, non-exclusive collective license for organizations training AI systems for uses external to their organization. It provides the rights to lawfully use a wide and diverse repertoire of content for training AI systems, as well as the subsequent use of the trained system as a service

AI Systems Training License: overview sheet Copyright Clearance Center

CLA: "training and fine tuning" (#cla-uses). Kopinor: #kopinor-scope. NDP: #sc-uses-ndp.

Our reading #ahead-fora-no-held

A pool on FORA licenses what leaves a fetch record, and a model trained on a copy it already holds leaves none.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

See #sc-held-fora.

A vote

Primary source #ahead-dpcmo-assembly

DPCMO's members vote in a general assembly.

Den øverste myndighed i DPCMO er generalforsamlingen. Medlemmerne er repræsenteret på generalforsamlingen ved en repræsentation pr. medievirksomhed.

Gennemsigtighedsrapport 2025 (transparency report, in Danish) DPCMO page 4

Translation: "The highest authority in DPCMO is the general assembly. The members are represented at the general assembly with one representation per media company."

FORA leaves governance to each operator.

FORA project statement FORA project no public URL

The protocol defines no governance for a pool.

Excluding one AI company

Primary source #ahead-fora-pool-term

On FORA, the operator admits AI companies to the pool, and a publisher's pool term covers each one it admits.

Scope-gating: the Exchange presents an offer for this term only when the scopes in the requester's signed request (Requester.scopes) cover ALL of these scopes (AND-semantics). Empty = public. A subscription term is Pricing{model:FREE} + scopes:["subscription:..."].

Protocol reference: FORA v1 FORA documentation LicenseTerm.scopes

The operator grants the pool's scope to the AI companies it admits; the publisher's pool term is presented to any agent holding it.

Back to the topics

Questions publishers ask

Primary source #faq-cost

FORA is free; each collective chooses how it prices its service.

FORA is free. Each collective chooses how it prices its service.

FORA project statement FORA project no public URL

The FORA project's statement.

Primary source #faq-keep-one-out

A publisher keeps one AI company out by scoping its terms to the AI companies it licenses; any it leaves out get no offer.

For a term the Exchange gates behind a delegation, it presents the term only when a verified delegation grants those scopes

Licensing terms FORA documentation Licensing terms, opening section

A term is presented only to requesters whose verified delegation grants its scopes, so an AI company outside those scopes gets no offer for it.

Primary source #faq-money

The operator collects the fees and pays publishers through its own payment provider; the protocol signs and records, and never moves money.

FORA never touches money. The protocol handles authorization and logging, while actual settlement happens outside the protocol through existing billing infrastructure.

How money flows FORA documentation "The Billing Chain"

Primary source #faq-ilc-allows

ILC allows a publisher to join a collective on FORA as well.

Is participation exclusive? No. Participation is non-exclusive. You keep the right to license your content directly to anyone, and any existing licensing deals remain unchanged.

Frequently asked questions Internet Licensing Collective "Is participation exclusive?"

Our reading #faq-exclude-first

If you want an ILC licensee to pay through a FORA collective, exclude it in ILC first.

Publisher may elect not to participate in any Collective License Agreement with a particular AI Company, or may withdraw some or all of its Designated Content from that AI Company’s Repertoire, at any time and for any reason, by authenticated notice through the Platform.

Publisher Participation Agreement, version 1.0, shown in ILC's publisher portal Internet Licensing Collective §2.4 no public URL

The portal check requires a declaration that grants every AI use (#ilc-uses-portal), so the same AI company would otherwise hold a license for the same uses twice. Exclusion in ILC is per AI company.

Our reading #faq-no-training

A collective on FORA doesn't license training: the pool licenses uses that start with a fetch, and a model that trains on a copy it holds fetches nothing.

REQUIRED. Hex-encoded detached Ed25519 signature over the canonical serialization of the ENTIRE Offer — every field, including `pricing`, `terms` (the full licensing payload), `expires_at`, and `exchange`.

Protocol reference: FORA v1 FORA documentation Offer.signature

The article's design for a pool; see #sc-held-fora.

Primary source #faq-own-training

You can still license training on your own priced term.

A resource carries zero or more `LicenseTerm` entries, at most 32. Each term is a complete, self-contained access arrangement, and a resource with several terms is the normal case.

Licensing terms FORA documentation "Overview"

A term's use restriction can permit training ("What the agent may do with the resource", "Restrictions"); a priced term sits beside the pool term on the same resource.

Primary source #faq-why-join

An AI company outside the collective meets a per-fetch price or a block at each member's pages.

A publisher's edge that refuses a request for licensed content carrying no valid signed URL answers `403` and SHOULD send `X-Content-Rules`, the absolute URL of the publisher's manifest

Protocol reference: FORA v1 FORA documentation "Edge discovery headers"

The manifest lists the Exchanges that sell the content. One license to many publishers and one report format are the collective's design, as the article describes it.

Our reading #faq-no-law

No law makes an AI company join a collective.

CCC’s AI Systems Training License is a voluntary, non-exclusive collective license for organizations training AI systems for uses external to their organization. It provides the rights to lawfully use a wide and diverse repertoire of content for training AI systems, as well as the subsequent use of the trained system as a service

AI Systems Training License: overview sheet Copyright Clearance Center

The collective AI licenses in this field are voluntary: CCC's is "voluntary", and PLS says "The scheme is entirely voluntary and opt-in". India's committee has only proposed a mandatory license (#w-dpiit). Unlike the US compulsory license for music (see /resources/music-dsp/), no statute found compels an AI company to take one.

Back to the topics