Free tools Windows power users keep installed
One-click scans. No signup required.
Publishers can license content to AI companies through a direct contract, a collective licensing service, or machine-readable web terms. The safest approach is to treat each use—model training, fine-tuning, retrieval-augmented generation (RAG), AI search, text and data mining, and use inside a publisher product—as a separate right. First establish who controls those rights, then negotiate the corpus, purpose, term, payment, reporting, attribution, security and termination provisions.
What an AI-content licence actually grants
“AI rights” are not one permission. A developer may want to copy an archive to train a foundation model, periodically retrieve articles for a RAG answer, index pages for an AI-search product, or use licensed material in a publisher’s own software. Those activities create different risks, technical requirements and value. A contract should name each permitted activity instead of granting an undefined right to “use content for AI.”
![Drafting Contracts: How and Why Lawyers Do What They Do [Connected Ebook] (Aspen Coursebook) (Aspen Coursebook Series)](https://m.media-amazon.com/images/I/51uJ-x2h7CL._SL160_.jpg)
As an Amazon Associate I earn from qualifying purchases.
Model training
Training usually involves ingesting a defined corpus into a process that changes model parameters. Ask whether the developer may retain the source files, make backup copies, or use the resulting model for products outside the named service. Because a trained model can continue to produce behavior after the source files are deleted, the agreement should address what happens to trained systems when the licence ends.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchFine-tuning
Fine-tuning adapts an existing model to a narrower corpus or task. It may require a different data set, term and payment from initial training. Specify whether the fine-tuned model can be reused for other customers, combined with third-party data, or transferred to another company.
#1 Best Overall
![Problems in Contract Law: Cases and Materials [Connected eBook with Study Center] (Aspen Casebook)](https://m.media-amazon.com/images/I/41GXyreL9oL._SL160_.jpg)
- Updated Contract Law Cases: Five new principal cases reflecting recent advances and improved statements
- Restored Classic Case: Oppenheimer & Co. v. Oppenheim for foundational perspectives
- New Review Options: Twelve fresh problems, including shorter ones, for varied teaching and contemporary fact patterns
- Enhanced Learning Tools: Eight new tables and flow charts for complex legal subjects
- Streamlined Notes and Text: Editing for conciseness without sacrificing coverage and incorporating new legal developments
Retrieval-augmented generation (RAG)
In a RAG system, the application retrieves passages at query time and supplies them to a model to produce an answer. The content may remain in a searchable index rather than being incorporated into model weights, but copying, indexing, caching and displaying passages still need permission. The Publishers Association reports that RAG licensing is growing and that these deals are often usage-based with regular payments; it does not publish a standard rate.
AI search and answer interfaces
An AI-search service may crawl, index, summarize or quote pages in response to user queries. Define whether the developer may show full text, excerpts, links, headlines, images, metadata or only statistical signals. RSL 1.0, an open standard announced on December 10, 2025, adds granular controls for AI-search uses, but a machine-readable declaration is not itself a negotiated payment guarantee.
Text and data mining
Commercial text and data mining (TDM) licensing predates generative AI by about a decade. A TDM grant can cover analytical copying without permitting public answers, model training or redistribution. State the exact outputs allowed and whether derived data may be retained after the licence expires.
Recommended Free Tools
Use in a publisher’s own product
A publisher may license its archive to an AI developer that powers an internal research tool, recommendation engine or customer-facing assistant. This can involve a limited field of use, support obligations and user-volume limits rather than a broad archive licence. Name the product, territory, user class and interfaces covered.
Rank #2
Three routes to market
| Route | When it fits | Rights and control | Payment and reporting | Important caveat |
|---|---|---|---|---|
| Direct licence | A publisher has valuable archives, a reliable feed or a specific strategic partner. | Terms can cover training, fine-tuning, RAG, search or proprietary products individually; parties can negotiate exclusivity, territory, security and sublicensing. | Fixed fees, usage-based fees, advances, revenue shares or hybrids are possible. Reporting and audit rights are negotiated. | Negotiation, diligence and enforcement are the publisher’s responsibility. Public deal prices are usually confidential. |
| Collective licence | Rights holders want pooled administration or lack the scale to negotiate with every developer. | A collecting body aggregates participating rights and sets the scope, opt-in rules and permitted uses. | The body may administer usage reports and distribute payments under its rules. | Coverage, launch status, eligible works, payment formula and opt-in terms must be confirmed for the relevant service. |
| Machine-readable web terms | A site wants to state permissions and compensation preferences consistently for crawlers and agents. | RSL can express free access, attribution, subscription, pay-per-crawl and pay-per-inference terms, including more granular AI-search controls. | The standard communicates a requested commercial model; it does not establish that a particular service will comply or what revenue a site will receive. | Implementation and enforcement by each AI service remain separate questions. RSL Collective membership is described as free and non-exclusive. |
What the market evidence shows—and does not show
The Publishers Association’s 2026 member survey of UK book and journal publishers found that ten responding publishers had already agreed one or more AI licensing deals. The association expected eight additional publishers to enter deals during 2026; that figure is a forecast, not a final count. The survey says the first AI-training licences began as early as 2023 and that most publishers making their first deals did so in 2024. Reported deal uses include training, fine-tuning, RAG and incorporation of other publishers’ content into proprietary products.
Those figures describe the surveyed UK publishing segment, not every country or type of publisher. The association says confidentiality prevents publication of an aggregated industry licensing value. No independently established universal price, royalty rate or guaranteed revenue figure should be assumed.
CLA announced work with Publishers’ Licensing Services (PLS) and the Authors’ Licensing and Collecting Society (ALCS) on a Generative AI Training Licence. CLA described coverage for text-based content used in training, fine-tuning and RAG and announced a target launch in the third quarter of 2025. The announcement reviewed here does not confirm that the licence is currently open, which works it covers, or its payment and opt-in terms. Treat it as a proposed route until the current service rules are verified.
CLA chief executive Mat Pfleger stated: “Training AI models on copyrighted content requires permission and compensation.” That principle does not answer which entity owns a particular work or what a specific licence costs; those questions remain contractual.
Establish who can grant the rights
Before discussing price, map ownership for every part of the proposed corpus. A publisher may own employee-created articles but hold only a licence for freelance material, photographs, illustrations, translations, syndicated stories, databases or previously published books. Check whether each author agreement is exclusive or non-exclusive, whether it is an all-rights transfer or a limited licence, and whether subsidiary rights include AI uses.
The Authors Guild advises authors to negotiate AI rights expressly. Its model language distinguishes training, RAG summaries and other interactive uses and states that rights not expressly granted remain reserved. A publisher should therefore identify any author consent, notice or revenue-sharing obligation before promising an AI company access.
Review whether the publisher may sublicense. An agreement that permits print, ebook or ordinary digital distribution may not authorize a sublicense for model training or retrieval. If a collective body will administer rights, confirm that the publisher can opt in the relevant works and that contributors’ rights are represented.
Decide how income is accounted for. The Authors Guild notes that, unless the parties agree otherwise, AI-licensing income may be applied against an author’s advance. The governing publishing agreement and applicable law control; a general industry assumption is not a substitute for that review.
A step-by-step licensing process
- Inventory the corpus. List titles, articles, editions, languages, media, publication dates and territories. Record exclusions such as third-party photographs or material under a non-transferable licence. Preserve a versioned manifest so both sides can identify what was delivered.
- Classify each requested use. Put training, fine-tuning, TDM analysis, RAG retrieval, AI-search indexing, display of excerpts and use in proprietary products into separate rows. For each row, state whether copying, caching, transformation, public display and human review are allowed.
- Choose a route. Solicit a direct proposal when you need bespoke controls or have a strategic partner. Investigate a collective when administration across many developers is more important than individual negotiation. Use machine-readable terms to communicate web preferences, while recognizing that a direct or collective contract is needed for enforceable commercial commitments.
- Prepare a data package. Define file formats, metadata, update frequency, correction procedures, access credentials and a contact for takedown or security incidents. Separate current feeds from historical archives and price them separately if their value or operational burden differs.
- Negotiate the grant. State the permitted purposes, models, products, users, territory, duration, exclusivity and sublicensing rules in plain language. Reserve every use not expressly granted. Include a process for adding new uses by written amendment.
- Set economics and measurement. Choose a fixed fee, usage metric, recurring subscription, revenue share or hybrid. For usage pricing, define the meter—documents ingested, tokens processed, retrievals, impressions, queries, active users or another verifiable unit—and the reporting period. If an advance is paid, specify whether it is recoupable against later royalties or treated separately.
- Build attribution and provenance requirements. Decide when the publisher and individual authors must be named, how source links are displayed, and what metadata travels with a passage. Require the developer to distinguish licensed material from generated text where that distinction matters to users.
- Operate and audit. Require periodic usage statements, a named reporting contact and records retained long enough to verify invoices. Negotiate a reasonable audit right, confidentiality protections for both sides and a remedy for under-reporting.
- Plan exit and changes. Address termination for breach, convenience and non-payment; deletion or return of source files; treatment of indexes, embeddings, caches, derived data and model checkpoints; and the handling of content removed for legal or factual reasons. Specify whether existing users may continue to access outputs and for how long.
Contract terms that prevent common disputes
Scope, versions and updates
Attach a corpus schedule or an API description. Identify version dates, permitted refreshes, corrections and whether a live feed is included. A right to use “the publisher’s content” without a schedule can silently expand as the archive grows.
Purpose and field of use
Use separate clauses for training, fine-tuning, RAG, search and internal tools. Prohibit using the corpus to train unrelated models, sell a competing database or create a derivative dataset unless those activities are expressly priced and permitted.
Term, territory and exclusivity
State start and end dates, countries, language editions and any embargo period. If exclusivity is requested, limit it by use, product, territory and time; an exclusive training grant should not automatically block a separate RAG or search arrangement.
Security, retention and confidentiality
Specify encryption, access controls, subcontractor obligations, incident notice and retention periods. Require a list of subprocessors and approval or notice rights for material changes. Confidentiality should protect both unpublished content and the AI company’s non-public technical information.
Derived data and outputs
Define ownership and permitted use of embeddings, indexes, annotations, evaluation sets, summaries and user prompts. Clarify whether the developer may retain aggregate statistics and whether generated answers may be used to build another dataset.
Attribution, accuracy and complaints
Set formatting for bylines, publication names and links. Establish a channel for correcting an outdated passage, removing a work or responding to an author complaint. Do not assume that an attribution promise fixes every copyright or defamation risk; allocate responsibilities and remedies explicitly.
Can a publisher get paid when an AI crawls its website?
Potentially, but payment depends on an enforceable arrangement and the service’s actual behavior. RSL is designed to let a publisher declare terms such as free access, attribution, subscription, pay-per-crawl or pay-per-inference in a machine-readable form. RSL 1.0 also adds controls for AI-search uses and a contribution option for noncommercial content or data. The RSL Collective says participation is free and non-exclusive.
The RSL materials do not establish that every crawler will honor a declaration, that joining produces income, or that a particular site has implemented the standard. A publisher seeking guaranteed compensation should negotiate directly or use a collective service whose current rules, coverage and enforcement are documented. Treat protocol publication as communicating terms, not as proof of acceptance.
An RSL announcement in December 2025 reported support from more than 1,500 media organizations, brands and technology companies. That is an endorsement figure reported by the RSL Technical Steering Committee; it does not show that all endorsers have implemented the standard or transact under it.
How to choose among the routes
- Choose direct licensing when you need control over a named model or product, have a clearly documented corpus and can support reporting and negotiation.
- Consider collective licensing when many rights holders need a common administrator, but verify that the service is live, your works are eligible, the permitted uses match your goals and the distribution formula is acceptable.
- Publish machine-readable terms when you operate a website and want a consistent signal to crawlers and agents. Pair it with technical monitoring and a contract if payment, attribution or enforcement is essential.
Compare options on rights coverage, corpus and content type, exclusivity, payment basis, attribution, reporting, sublicensing control, implementation burden, enforcement, geographic scope and the public availability of current terms. No route is universally best; the right choice depends on rights ownership, bargaining power, technical capacity and the use being licensed.
Questions to resolve before signing
- Who owns or controls every item, and can that party authorize AI training, retrieval and sublicensing?
- Does the grant cover only named models and products, or future systems as well?
- What is the measurable billing event, and can the publisher verify it?
- How are author and contributor shares calculated, and can income be credited against an advance?
- What happens to copies, indexes, embeddings and trained systems after expiry or termination?
- Can the developer combine the corpus with other data or permit affiliates and contractors to use it?
- What attribution, provenance, correction and takedown process will users see?
- Which laws govern the agreement, and where can disputes be brought?
Bottom line
Licensing to AI companies is a real and developing market, but it has no published universal price or standard contract. Separate every AI use, verify the chain of rights, choose direct, collective or protocol-based terms for a defined business reason, and insist on measurable reporting and an exit plan. Current collective-service availability and implementation details can change, so confirm them before committing your catalogue.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
