Geoff CuddA course or publishing business sells its data to AI by licensing the long-form content it holds the training rights to, to a buyer like Mercor.
Mercor pays publishers and education companies $35,000 to $370,000. It wants hundreds of titles or hours from a company of 10 to 500 people, in any language.
One question decides everything: do you hold the AI training rights, or does your author? Most contracts written before 2023 never say.
For the full process across every industry, start with our guide to how to sell your company’s data to AI companies.
Key takeaways
- Mercor’s publisher and education listing pays $35,000 to $370,000 for full texts, course material and video with transcripts.
- It rules out content distributed under someone else’s license and content that is already open access.
- Short articles alone aren’t enough. It wants long-form text, courses or video, in the hundreds of titles or hours.
- HarperCollins’ AI deal pays $5,000 a title, split 50-50 with the author, and authors have to opt in.
- Wiley booked $49 million in AI revenue in fiscal 2026, and over $110 million since it started.
- The Authors Guild says AI training rights sit with the author unless the author granted them.
- Student records are a dead end. Under FERPA, a party that receives them can’t pass them on.
What course and publishing data do AI companies buy?
Mercor’s publisher and education listing, read November 9, 2026, splits it like this:
| Content | Mercor’s view | Main risk |
|---|---|---|
| Full texts of books and journals | Core of the license | Author contracts that never mention AI |
| Course material | Core of the license | Guest instructors and licensed curriculum |
| Video with transcripts | Core of the license | Music beds, stock footage and on-camera releases |
| Recent research papers that aren’t open access | Adds to what you earn | Author addenda and funder mandates |
| STEM textbooks and lecture recordings | Adds to what you earn | Figures and images licensed in |
| Medical and science educational video | Adds to what you earn | Patient images and consent forms |
| Fiction, essays and teaching writing | Adds to what you earn | Rights reverted to the author |
| Content in several languages | Adds to what you earn | Translator rights |
| Content under someone else’s license | Not a fit | Rejected outright |
| Content that is already open access | Not a fit | Rejected outright |
Mercor names the companies it wants: academic and professional publishers, textbook and courseware companies, e-learning platforms, medical education companies, and magazine and literary publishers.
It works with exports from a CMS, an LMS, or an XML and EPUB pipeline.
Its four opening questions are the whole diligence in miniature. What do you hold the training rights to, how many titles or hours, which subjects and languages, and is any of it licensed for AI training already?
How much is course and publishing content worth to AI?
The published deals give you a per-title benchmark instead.
| Deal | What it paid | Terms |
|---|---|---|
| Mercor publisher listing | $35,000 to $370,000 | Whole catalog, scoped on a call |
| Wiley, fiscal 2026 | $49 million in AI revenue | Up 23%, over $110 million since it started |
| Taylor & Francis with Microsoft | Nearly £8 million ($10 million) | First year, non-exclusive |
| HarperCollins AI deal | $5,000 a title | Split 50-50 with the author, opt-in |
| Anthropic’s Bartz settlement | About $3,100 a work | A court settlement, not a license |
Wiley’s own fiscal 2026 results put AI revenue at $49 million for the year to April 30, 2026, up 23%.
Taylor & Francis sold Microsoft access to its authors’ research for nearly £8 million in the first year, The Bookseller reported in 2024. Academics said they were never told.
Those are catalogs of tens of thousands of titles. A course business with 400 hours of video sits at the Mercor end of the table, not the Wiley end.
Mercor also shows a $50,000 referral fee on this listing. That goes to whoever refers the company, not to the company.
Our guide to how much company data is worth to AI compares published payouts across industries.
Do you hold the AI training rights?
Work through the catalog in four questions.
- What does the grant of rights say? A broad grant of “all rights in all media now known or hereafter devised” is one thing. A grant of print and electronic book rights is another.
- Have any rights reverted? Out-of-print clauses hand rights back to the author, often automatically.
- Who else is in the file? Translators, illustrators, photographers, figure permissions and licensed-in chapters each carry their own grant.
- Is it already licensed? Mercor asks, and an exclusive deal with one buyer ends the conversation with the rest.
The two piles Mercor refuses are worth naming. Content you distribute under someone else’s license isn’t yours, and content that’s already open access is free for anyone to train on, so a buyer won’t pay for it.
That second one surprises academic publishers. A gold open access paper under CC-BY has no scarcity left to sell.
17 U.S.C. 204(a) is the rule to keep in mind. If you can’t point to the signed page, you don’t have the right.
The Authors Guild’s statement on the HarperCollins deal puts it flatly: these rights belong to the author.
That deal is the template people copy, so the terms are worth knowing.
- $5,000 a title, split 50-50 between the author and the publisher
- Opt-in: the author has to agree, title by title
- Payments separate from royalties, not set against the advance
- Output caps of no more than 200 consecutive words and 5% of a book’s text
- A pledge not to scrape text from piracy sites
The Guild accepted the shape of the deal and objected to the split, arguing the author should get more than half.
Taylor & Francis went the other way and licensed without telling its authors. The coverage that followed is the reason to put consent in writing first.
Course businesses face the same question with different names. Guest instructors, contract curriculum writers and anyone on camera all need a release that covers AI training, and a revenue share is the fastest way to get one.
Can you include student data?
Under 34 CFR 99.33(a), a school discloses education records only on the condition that the recipient won’t redisclose them without consent, and the recipient may use them only for the purpose of the original disclosure.
An edtech company that receives student records as a school official is that recipient. Licensing them to an AI buyer isn’t the purpose they were disclosed for.
Children’s data adds another layer. COPPA requires verifiable parental consent before you disclose personal information collected from a child under 13.
So split the export cleanly.
| Item | What to do |
|---|---|
| Lessons, scripts, slides and workbooks | Include; this is the content you own |
| Instructor video and transcripts | Include, with a release that covers AI training |
| Student assignments and graded work | Leave out |
| Course forum and Q&A threads | Leave out, or strip every learner identifier and quote |
| Recorded live sessions with learners on camera | Leave out unless every participant consented |
| Enrollment, grade and progress records | Leave out; FERPA and state student privacy laws |
| Support tickets and refund notes | Strip names and emails, or leave out |
Our guide to anonymizing data for AI training covers how to strip identifiers properly.
Who buys course and publishing content?
- Mercor: full texts, course material and video with transcripts, for companies of 10 to 500 people anywhere, with an established catalog. Pays $35,000 to $370,000.
- Created by Humans: a book rights platform for authors, agents and publishers. It offers training, reference and transformative rights, negotiates each agreement for your approval, and publishes no rates.
- The labs directly: Microsoft, OpenAI and Amazon have signed publisher deals, but at catalog scale. Wiley, Taylor & Francis, HarperCollins and News Corp are the public examples.
- micro1: buys operational data, not content. Its data partnerships page lists software, finance, support, sales, legal and operations, so it’s a fit for how your business runs, not your catalog. You can apply to micro1 here.
Our list of companies that buy business data for AI has each buyer’s terms.
What are the steps for a publisher or course business?
- Inventory the catalog. Titles, courses, hours of video, languages, and how recent each one is.
- Read the grant of rights. Title by title, looking for the words that cover a use nobody imagined when it was signed.
- Get consent where the author holds the right. An opt-in with a revenue share, following the HarperCollins pattern.
- Cut what won’t sell. Open access material, licensed-in chapters and anything already under an exclusive AI deal.
- Apply to a buyer. Mercor signs an NDA, scopes the catalog and runs the export.
- Separate the student data. Content in, learner records out, with the identifiers stripped from anything in between.
- Have counsel review the license. Our AI data licensing agreement checklist covers output limits, exclusivity and term.
- Do the per-title math. Divide the offer by titles and hours, then compare it with the $5,000 a title HarperCollins pays.
Ask for the output caps by name. Limiting what the model can reproduce verbatim is how HarperCollins protected the value of the books it licensed, and it costs a buyer nothing to agree.
Common mistakes when selling course and publishing data to AI
The most common mistake is assuming the publisher holds every right in the contract. These are the others:
- Reading a grant of print and ebook rights as covering AI training
- Licensing without telling the authors, which is what made the Taylor & Francis deal a news story
- Offering open access papers, which have nothing left to sell
- Including licensed-in chapters, figures and translations
- Forgetting that out-of-print clauses may have reverted rights already
- Putting student forum threads in the export because they look like content
- Shipping short articles alone, when the listing wants long-form
- Signing without output caps on verbatim reproduction
- Reading the $50,000 referral fee as what the company gets paid
Frequently asked questions
Can you sell online course content to AI companies?
Yes, if you hold the rights. Mercor’s publisher and education listing covers course material and video with transcripts, takes companies of 10 to 500 people, and pays $35,000 to $370,000.
How much do AI companies pay for books and courses?
Mercor publishes $35,000 to $370,000 for a catalog. Per title, the HarperCollins deal pays $5,000, split 50-50 with the author. Wiley booked $49 million in AI revenue across fiscal 2026.
Who owns the AI training rights, the publisher or the author?
Whoever the signed contract says, and the Authors Guild reads standard trade contracts as reserving these rights to the author. A transfer of copyright needs a signed writing under 17 U.S.C. 204(a).
Do you need author permission to license books for AI training?
In most cases yes. The HarperCollins deal is opt-in, title by title, and splits the $5,000 fee 50-50, which the Authors Guild treats as evidence that the right sits with the author.
Can you include student data in an AI data deal?
No. Under 34 CFR 99.33(a), a party that receives education records can’t redisclose them without consent and may use them only for the original purpose, and COPPA adds parental consent for children under 13.
Does open access content have any value to AI buyers?
No. Mercor lists content that is already open access as not a fit, because anyone can train on it without paying. Recent material behind a paywall is what buyers want.
Last updated: November 9, 2026
Disclosure: some links on this page are affiliate links. If you sign up through one, we earn a commission at no extra cost to you.
This isn’t legal advice. Have counsel review your author contracts, instructor releases and any data license before you sign.
