A publisher can license a catalogue for AI training only where it holds the rights. AI buyers ask for long-form content such as journals, textbooks, courses and lecture video, and they value recent work that is not open access. The deciding question is whether your author, contributor and permissions contracts reach AI training. Work made for hire and full copyright assignments usually do. Older trade contracts and licences to publish often do not, and those titles need the author to agree first.
- The right to license belongs to whoever holds the copyright or an express grant that covers the use.
- A journal or anthology contribution with no written transfer gives the publisher only a narrow reprint privilege.
- Open access content and content you distribute under someone else's licence are not what buyers pay for.
- Sort the list by rights status before you count it. Only the clear bucket goes into a first conversation.
Can a publisher license books and journals for AI training?
Yes, for the works it holds the rights to. Copyright law gives the owner "the exclusive rights to do and to authorize" copying and the preparation of derivative works (17 U.S.C. 106). A publisher holds those rights only as far as its contracts with authors and contributors gave them to it. So the answer is different for each kind of contract on the list.
The table sorts a catalogue the way a rights manager would. This page is general information, not legal advice. Contracts differ, so read your own.
| How the publisher holds the work | What the publisher usually has | Can it grant an AI training licence? |
|---|---|---|
| Work made for hire: staff writers, or commissioned instructional texts and tests under a signed agreement | The publisher is treated as the author | Usually yes. Check third-party material inside the work |
| Full copyright assignment in writing (common for older journal articles and some academic books) | All the exclusive rights | Usually yes. Check what the contract promised the author about reuse |
| Exclusive licence to publish with listed rights (most trade books) | Only the formats, territories and subsidiary rights listed | Only if the grant reaches the use. Otherwise the author has to agree |
| Journal or anthology contribution with no written transfer | The privilege of printing it in that collective work and its revisions | No, not without the contributor |
| Open access under a Creative Commons licence | Nothing exclusive. Anyone may reuse the work on the licence terms | There is little to license |
| Images, excerpts and lists you distribute for someone else | Permission for a named use | No. The licensor decides |
What do AI buyers ask publishers for?
Long-form content with clear rights. According to one buyer's public listings, read 15 November 2026, a buyer looks for "content the company holds the rights to license for AI training", "long-form text, courses or video, not short articles alone", and "hundreds of titles or hours". The licence covers full texts, course material and video with transcripts.
The same listing says the most valuable material is recent content that is not open access, across science, medicine, law, finance and the humanities, along with educational video and lecture recordings.
Form matters as much as subject. Buyers work from structured source files, not scans. EPUB is the W3C format that "defines a distribution and interchange format for digital publications and documents" (W3C). Journal publishers usually hold articles as XML in the Journal Article Tag Suite, which grew out of work at the U.S. National Library of Medicine (JATS). Your title metadata is probably already in ONIX, "the international standard for representing and communicating book industry product information" (EDItEUR). Those three exports answer most of a buyer's first questions.
Which publishers and education companies qualify?
The published fit is a company of 10 to 500 employees with an established catalogue, in any country, with content in more than one language welcome (one buyer's public listings, read 15 November 2026). The kinds of company named are academic and professional publishers, textbook and courseware companies, e-learning platforms, medical education companies, and magazine and literary publishers.
Two things are listed as not a fit: content distributed under someone else's licence, and content that is already open access. A distributor, an aggregator or a press whose list is mostly licensed in from other houses has less to offer than its title count suggests.
Course libraries and training video follow the same rules with one more layer, the releases signed by the people on camera. The guide to licensing training course content covers that.
What about third-party material and open access?
A book you own can still contain material you do not. Photographs, figures, maps, long quotations and reprinted chapters usually came in under a permission for a named edition, format or print run. The permissions log decides whether those items can go into a licensed set or have to be stripped from the files first.
Open access is different. Under the CC BY licence anyone is free to "copy and redistribute the material in any medium or format for any purpose, even commercially" (Creative Commons). A buyer does not need you for that, which is why open access content is listed as not a fit. Hybrid journals should flag each article, because the paid and the open articles sit in the same volume.
The wider legal picture is unsettled, and it is one reason buyers pay for clear rights. The U.S. Copyright Office wrote that "voluntary licensing is already happening in some sectors, and it appears reasonable or likely to be developed in others" (Copyright Office report). The report is a pre-publication version and is not law.
What is a publishing catalogue worth to an AI buyer?
The one published range we can cite is $35,000 to $370,000 for the company, which the listing says "depends on catalog size and rights" (one buyer's public listings, read 15 November 2026). Read it as a ceiling for a strong catalogue, not a typical result. It is what one buyer advertises, not what any publisher has been paid.
Figures from the largest houses are a different scale and should not be used to price a small list. Wiley reported "$40 million in total AI licensing revenue realized in Fiscal 2025 compared to $23 million in Fiscal 2024" (Wiley, June 2025) and "$49 million of AI revenue" for fiscal 2026 (Wiley, June 2026). Those are the totals of one of the largest research publishers, across many agreements. The deals tracker lists other public agreements.
| Raises the value | Lowers the value |
|---|---|
| Recent research papers that are not open access | Content that is already open access |
| STEM textbooks and lecture recordings | Short articles with no long-form work behind them |
| Medical and science educational video with transcripts | Video with no transcript and no releases |
| Fiction, essays and teaching writing | Rights that cannot be shown title by title |
| Content in several languages | Scanned pages with no structured source files |
The left column is from the listing's own list of what adds to what the company earns. The right column is the reverse of the listing's fit criteria and our reading of them.
Example: a rights audit of a 600-title list
A fictional professional publisher with 35 staff has 600 books and two journals. Before talking to anyone, its rights manager sorts the list by contract type. The counts are invented to show the method.
| Bucket | Titles | Rights position | Decision |
|---|---|---|---|
| Commissioned handbooks under work-made-for-hire agreements | 210 | Publisher is the author; agreements on file | In the first set, after third-party figures are checked |
| Monographs with full written assignment | 140 | All rights assigned | In the first set |
| Trade titles with a listed grant of rights | 170 | Silent on AI | Opt-in letter to each author; in only if signed |
| Edited volumes with chapter contributors | 55 | Some contributor agreements missing | Held back until each chapter has a signed transfer |
| Titles licensed in from other publishers | 25 | Rights belong to the licensor | Out |
| Journal articles, subscription | 3,400 articles | Copyright transfer forms since 2009 | In from 2009; earlier volumes held back |
| Journal articles, open access | 900 articles | CC BY | Flagged; not part of the offer |
Fictional example. The first set is 350 books and 3,400 articles, well under the headline count, and it is the only number the publisher quotes.
How to prepare and count the catalogue
A buyer's first questions are how much, about what, in which languages and with what rights. One spreadsheet answers them. No content needs to leave the building to build it.
- Export the title list from your ONIX feed or title management system: identifier, title, subject code, publication year, language, page count.
- Add a rights column with three values: clear, needs the author, excluded. Add the contract type and where the signed document is filed.
- Add a word count per title from the source files. Words, not titles, are how text is sized.
- For journals, count articles by year and mark each one subscription or open access.
- For courses and video, count hours, and mark which have transcripts and signed releases.
- Flag titles with heavy third-party content such as art books, anthologies and heavily illustrated texts.
- Note anything already licensed for AI use, to whom, and whether that licence was exclusive.
- Record the file format per title: EPUB, XML, print-ready PDF or scan only.
What does a licence usually allow and forbid, and how long does it take?
A training licence is normally non-exclusive, limited to a named use and a named party, and for a fixed term. It usually forbids resale or redistribution of the files and sets rules about how much of a work a model may reproduce in its output. Whether the licence covers training, retrieval at answer time, or both changes the price and the risk, and the guide to training and retrieval licences explains the difference. The data licensing agreement guide lists the clauses to expect.
Publishers should add three points of their own: how income is shared with authors and how that is reported, whether the publisher or the author is credited, and what happens to the files and to anything built from them when the term ends.
No public source gives a reliable timetable for a data licence, so this page does not quote one. The order of work is fixed, though: rights audit, author opt-ins where needed, a sample set, terms, then delivery. The rights audit and the opt-in round are the slow steps, and both are in your hands before any buyer is involved.
Common mistakes publishers make
Most problems come from counting the list before sorting it.
- Quoting the whole backlist when only part of it has clear rights.
- Treating a journal's right to print an article as a right to license it for a new use.
- Assuming a commissioned work is work made for hire with no signed agreement on file.
- Leaving licensed-in photographs, figures and excerpts inside the files.
- Mixing open access and subscription articles in one count.
- Granting exclusivity over content that was already licensed elsewhere.
- Agreeing terms before deciding how authors will be paid and told.
A publishing catalogue as a data asset
A catalogue with clear rights, structured files and recent long-form work is the kind of set AI buyers look for from publishers and education companies. The value is in the part you can show title by title, so the rights spreadsheet is worth more than the headline count.
Origin works with buyers and helps a publisher describe that set, find out what it could be worth and decide whether to go further. Nothing is exported at the start. Other record types are covered in the guide to what company data can be licensed and the AI licensing overview. Origin Data Partners is an independent data partner that helps companies get their data bought. A buyer may pay Origin when a deal closes.
Does your company hold records like these? See what your data is worth.
Common questions
Can a publisher license books for AI training without asking authors?
Only where its contracts already give it the right: work made for hire, a full written assignment, or a grant of rights wide enough to cover the use. Where the contract is silent or narrow, the author has to agree first. Many publishers ask in every case.
Who holds AI training rights to a journal article?
The author, unless there is an express written transfer. Under 17 U.S.C. 201(c) a journal with no transfer has only the privilege of publishing the article in that journal, its revisions and later issues in the same series.
How much do AI companies pay publishers for a catalogue?
One buyer's public listing, read 15 November 2026, gives $35,000 to $370,000 for a publisher or education company, depending on catalogue size and rights. That is a published ceiling, not a typical result. Large research publishers report far bigger totals across many agreements.
Is open access content worth anything to an AI buyer?
Very little. A CC BY licence already lets anyone copy and reuse the work for any purpose, so a buyer does not need to pay the publisher for it. Buyers list open access content as not a fit.
Do course videos and lecture recordings count as licensable content?
Yes. Buyers ask for courses and video with transcripts alongside full texts, and they count it in hours. You also need releases from the people who appear in the recordings.
What file formats do AI buyers want from a publisher?
Structured source files: EPUB for books, XML such as JATS for journal articles, and the title metadata from your ONIX feed. Scanned pages with no text layer are worth much less.
Sources and further reading
- 17 U.S.C. 106: exclusive rights in copyrighted works
- 17 U.S.C. 201: ownership of copyright (works made for hire, contributions to collective works)
- 17 U.S.C. 204: execution of transfers of copyright ownership
- 17 U.S.C. 101: definitions (work made for hire, instructional text, collective work)
- U.S. Copyright Office: Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version, PDF)
- Authors Guild: statement on prior consent for AI use of academic and news content
- Authors Guild: AI-related model publishing contract clauses
- Creative Commons: Attribution 4.0 International (CC BY 4.0) deed
- Wiley: fiscal 2025 results press release (17 June 2025)
- Wiley: fourth quarter and fiscal 2026 results press release (16 June 2026)
- W3C: EPUB 3.3 Recommendation
- U.S. National Library of Medicine: JATS and the NLM DTDs
- EDItEUR: ONIX for Books, overview
