How to license books and publishing archives to AI companies

Books are the most contested training data. On July 20, 2026 a US federal court granted final approval to the $1.5 billion Bartz v. Anthropic settlement over books used without permission, which the Authors Guild puts at about $3,000 per work. Licensed, rights-cleared text is the alternative, and publishers with clear rights are being paid for it.

The US Copyright Office's report on generative AI training, released in pre-publication form on May 9, 2025, said licensing agreements for AI training "are fast emerging in certain sectors" and that government intervention "would be premature at this time". It also said that "making commercial use of vast troves of copyrighted works" to compete in existing markets "goes beyond established fair use boundaries".

What do AI companies pay for books?

AI companies have paid publishers per title, as upfront fees and as recurring payments. The reported figures come from the publishers' own disclosures and trade press.

The Bartz settlement figure is a damages settlement, not a license price, so treat it as context rather than a benchmark. The AI data deals tracker lists these deals with the news and archive licenses signed by OpenAI, Amazon, Meta and Google.

Who licenses book and publishing archives?

Book and publishing archives are licensed directly to AI developers, through rights-holder marketplaces and through training-data companies. Microsoft appears in two of the reported deals above, and Human Native was built to help creators and rights holders license content to AI developers before Techzine reported its acquisition by Cloudflare in January 2026.

See our profile of Human Native and our page for publishers and education companies, which covers editorial records and assessment items as well as text.

Do you hold the AI rights in your backlist?

You hold the AI rights in your backlist only where your contracts grant them. Under US law the author owns the copyright in a book unless it is a work made for hire, so a publisher holds only the rights the author contract transfers, and a contract granting print and ebook rights may say nothing about training.

Authors are pushing for explicit terms. The Authors Guild published a model clause on March 1, 2023 that says the publisher may not use the work for AI training, nor sublicense others to, "without Author's specific and express permission". Expect agents to ask for it, and audit contracts title by title.

Educational and reference publishers often have more work-made-for-hire material. The US Copyright Office's Circular 30 lists instructional texts, tests, answer material for tests, compilations, translations and contributions to collective works among the nine categories where a signed agreement can make the publisher the author.

A publishing archive is more valuable when the rights are documented title by title and the text is clean, structured and hard to find elsewhere.

How should authors share in AI licensing revenue?

How authors share in AI licensing revenue is a commercial choice where the contract is silent, and the reported deals show two approaches. Publishers Weekly reported that HarperCollins offered authors an opt-in and an equal split of $5,000 per title, while The Bookseller reported that Wiley confirmed authors could not opt out.

An opt-in with a clear split costs time but gives you a clean rights record for each title, which buyers check first. Take independent publishing and copyright advice before you choose.

How should you package book files?

You should package book files from production sources, with a per-title rights record and third-party material removed.

• EPUB or XML from production files, keeping chapter and heading structure
• A per-title manifest with ISBN, edition, rights basis and the date the rights were confirmed
• Third-party images, epigraphs, long quotations and permissions-cleared extracts removed
• Illustrations and figures supplied separately, only where you hold their rights

Our guide to exclusive and non-exclusive licenses explains how term and exclusivity affect the price of a backlist.

If you hold UK or EU data

The UK government's report on copyright and AI, published on March 18, 2026 under the Data (Use and Access) Act 2025, says a broad copyright exception with an opt-out is no longer its preferred approach and proposes not to intervene in the licensing market for now. Under section 11 of the Copyright, Designs and Patents Act 1988 the author is the first owner of copyright, so UK licensing also rests on contract.

How do you start licensing a publishing archive?

You start with a rights audit: list titles by contract date and template, mark which grant AI or sublicensing rights and which need a variation, and note third-party permissions. That audit sets the size of the licensable catalog.

Our seller vetting covers ownership, chain of title and third-party content before any buyer sees a sample. Book a free consultation to find out what your backlist could earn.