Updated
For publishers and course companies
Sell Publisher Data to AI Companies
Short answer
Yes, for content you hold the rights to license for AI training. Some AI data programs pay for long-form journals, textbooks, courses and video. The publisher keeps its catalog. Author contracts decide what you can include, so a lawyer should read them first.
To sell publisher data to AI companies, you license long-form content for AI training and keep your catalog. A license is a written permission with limits. It is not a sale of your books, your journals or your courses. Education companies with courses and lecture video can take this path too.
The hard part is rights. Owning a title on your list and holding the right to license it for AI are two separate things. This page covers the catalog traits programs ask about, the publishers they pass on and how to check author deals. DillisData charges publishers nothing. Our money comes from referral fees that programs pay.
- Long-form content wanted
- Author rights checked first
- Your catalog stays yours
What programs ask for
What do AI companies look for in publishers?
Programs start with the first four rows, and each program has its own rules. The rest can lift interest in a catalog. Content in many languages is welcome in some programs.
| What some programs ask for | Where it usually lives | Why they want it |
|---|---|---|
| Content your company holds the rights to license for AI training | Author, editor and contributor contracts | Without clean rights, a program can't safely use the work |
| Long-form text, courses or video, not short articles alone | Book files, Open Journal Systems, Canvas, Thinkific, Teachable | Long works hold full lines of reasoning, not just quick takes |
| Hundreds of titles or hours | An established catalog and backlist | A small list gives too little for a program to work with |
| A company of about 10 to 500 employees, based anywhere | Your payroll | One size rule we see for publishers. Others differ |
| Recent research papers that aren't open access | Journal archives, often tagged in JATS XML | Paywalled work is harder for a lab to get on its own |
| STEM textbooks and lecture recordings | Book production files, Kaltura or a similar video library | They teach a subject step by step, with worked problems |
| Medical and science video with transcripts | Video archives and captions | Video with accurate text gives two forms of the same lesson |
| Fiction, essays and teaching writing | Trade and academic backlists | Skilled long writing is what these programs name as a plus |
| Content in several languages | Translated editions and foreign lists | Good text in other languages is harder to find |
An honest filter
Which publishers are not a fit?
These usually won't clear the bar. It saves everyone time to say it here.
Content under someone else's license
Maybe you distribute a title for another publisher, or you license it in. Then it isn't yours to pass along for AI use.
Open access catalogs
Content that is already open access can often be reached for free. Some programs pass on it. Creative Commons licenses also carry their own conditions, like credit or no commercial use.
Short articles alone
News briefs, blog posts and short web pieces don't meet the long-form rule by themselves.
Very small lists
Programs that use this rule look for hundreds of titles or hours. A few dozen books is likely too few for them.
Companies that only hold user uploads
Authors or instructors may post their own work on your platform and keep their rights. Then you may have nothing to license.
Size up your catalog in two minutes
The short quiz asks about your team, years in business and what you hold. You send no manuscripts and no contracts.
Author contracts
Who holds the AI rights to your titles?
Start with what the law says about rights. US copyright law (Title 17, chapter 2) lets any right in a copyright be split off and owned alone. A transfer of ownership is only valid if it is in writing and signed. So your rights are whatever your contracts gave you, and no more.
Older contracts rarely mention AI, because AI training didn't exist when they were signed. Language like "all media now known or later devised" may or may not reach it. No one should guess. Ask your lawyer to read your main contract forms, then sort your list into cleared, unclear and not yours.
Journals and anthologies need their own check. Section 201(c) says each contribution has its own copyright, which starts with its author. Without an express transfer, the law presumes the publisher holds limited rights. They cover use in that work, its revisions and later works in the series. So signed transfer forms matter a great deal.
Backlists have one more wrinkle. Section 203 lets authors end many grants after about 35 years, under set conditions. It doesn't reach works made for hire. Staff writing is usually in that group, so course content your employees wrote may be cleaner to license.
Asking authors is a real option. Publishers Weekly reported on one HarperCollins AI deal. It ran three years, needed authors to opt in and split the fee evenly with them. That is one reported deal and not a rule.
The U.S. Copyright Office's report on AI training says voluntary licensing is already happening in some fields. It adds that availability so far is uneven. It also expects some AI training to qualify as fair use and some not to. The report is a pre-publication version. It does not settle what your contracts allow, so ask your lawyer.
Keep out
What should a publisher leave out?
Your team picks the titles. The contract should list what stays out. This is not legal advice, so have counsel review it.
Permissions you only cleared for print
Textbooks are full of photos, charts and quotes used with permission. Those permissions may not cover AI use. Mark them, or leave those titles out.
Stock media inside courses
Lecture video may use licensed music, clips or images. Those licenses belong to other people, so check them first.
Instructors who kept their rights
Some teachers own their course content under contract. Read each agreement before a course goes in.
Student records
Keep grades, names and student posts out. Under FERPA rules, schools generally need signed consent to share identifying student records. If you hold records for a school, ask your lawyer.
Reviewer files and unpublished drafts
Peer review reports and rejected manuscripts came to you in confidence. Authors never agreed to reuse. Leave them out unless a lawyer says otherwise.
Other people's names in recordings
Lectures can catch students on camera or on the mic. Programs say they remove personal details, though no tool catches it all. Ask to review a sample.
How it works
How does a catalog license come together?
We connect publishers with programs and stay out of the content. For the broader picture, see how businesses sell data to AI companies.
Take the quiz
Answer a few questions on staff, years in business and your catalog.
2 minutesRead your estimate and three best matches
We set your answers next to what each program publishes, then show why each one fits. Your estimate is a guide, and it is not an offer.
Audit your rights
Your lawyer reads the contract forms. You sort titles and series into in or out.
Sign with your chosen program
The deal is directly between you and the program. For some programs, a representative reaches out to you instead of a sign-up link.
Deliver files to the program
Files go from your archive straight to the program. We do not take custody of them.
Public filings
What have publishers reported from AI licensing?
These come from large public companies and have nothing to do with DillisData. They say little about what a smaller list can earn. They do show the money is real.
AI revenue Wiley reported for its 2026 fiscal year, up 23%. Its lifetime AI revenue passed $110 million.
Wiley 8-K, Jun 2026Data access revenue in 2024 at Taylor & Francis, in part from AI companies. Its owner called this revenue non-recurring, so it can come in bumps.
Informa results, Mar 2025The first fee Microsoft agreed to pay Informa for data access, plus payments over three more years.
Informa update, May 2024Timing
Why start a rights audit now?
No one can say how long programs will keep paying at these levels. Here is what makes waiting costly for a publisher.
Authors get harder to find
Contacts go stale, agents retire and estates change hands. Getting a yes on an old title gets harder over time.
Other deals may shape your options
Some publishers are already signing. A content type a program can buy from others may draw less interest from it.
Contracts can be fixed going forward
A rights audit shows gaps in your current forms. Your new author deals can then address AI use in plain terms.
Old files decay
Early book files, tapes and old course videos get lost or become hard to open. Sort them while someone remembers them.
Know your number before you call your authors
An estimate tells you if a full rights audit is worth doing. The quiz is short and needs no files.
Questions
Questions owners ask
Do we need every author's permission?
It depends on what each contract gave you. Some titles may be clear, and others may need a yes from the author. Your lawyer should read your contract forms and tell you which is which.
Our contracts say all rights. Is that enough for AI?
It may be, but no one should assume so. Wording and dates matter. Ask your lawyer to review your main forms.
Isn't AI training just fair use, so why would anyone pay?
The Copyright Office expects some uses to qualify and some not to. Programs that want clean rights still pay for a license. A license can also bring files and metadata that scraping can't.
We sell online courses, not books. Do we count?
Yes, if you hold the rights. Courses and lecture video count as long-form content. Check any instructor contracts and licensed media first.
Can we license only part of our list?
Yes. You choose the titles, series or years that go in. A smaller set may mean less money, and it must still meet a program's size rule.
Can we keep selling the same books?
Yes. Your copies, sales and distribution carry on as they are. The license only lets a program use a copy, under the terms you sign. Your lawyer should read those terms closely.
What might a publisher be paid?
What programs typically pay overall is $100K to $5M+, largely depending on publisher size. That is no promise. Deals reported by big publishers don't predict yours, so use your estimate instead.
Does DillisData see our manuscripts?
No. You answer a quiz and we never touch your files. The publisher signs with the program itself, and files move straight there.
Free estimate
Find out what your catalog could earn
Two minutes of questions. You see a careful number and three matched programs, with the reasons for each.
No manuscripts or contracts needed.