DataPacts

Course and Education Content as AI Training Data

Checked

At a glance

Course and education content, meaning course videos, lecture recordings, textbooks, courseware and the question banks and learner work around them, can be licensed for AI training only by a holder who owns the training right in it, and a course often has more than one possible holder: the instructor, the institution and the platform that hosts it.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

As of November 17, 2026, the two education sites whose terms could be read state a position on AI training: Coursera bars training AI models on its content without prior written consent, and MIT OpenCourseWare permits it only for non-commercial models.

Wiley states USD 16.0 million of Learning AI licensing revenue for its fiscal year to April 30, 2026, and no other education publisher's terms or values are on record. This page follows AI training data licensing through ownership, platform terms, student data limits and the publisher record.

The figures and quotations here were read from the named sources on the dates given; how we checked this page sets out the method.

What Kinds of Course and Education Content AI Buyers Ask For

Of the education content types on this page, Mercor's publisher or education listing names textbooks, courses and educational video, and the sources read name no program that asks for assessments or tutoring transcripts.

The table compares five content types by who usually decides licensing and what a buyer on record says, as of November 17, 2026.

Content type Who usually decides licensing What a buyer or source on record says Source
Course videos and lecture recordings the author or employer; a platform agreement if hosted Mercor: "educational video and lecture recordings", "Video with transcripts" api.data.mercor.com listing publisher
Textbooks and courseware the publisher or author under the publishing contract Mercor: "STEM textbooks", "Full texts", "Course material"; "Hundreds of titles or hours" api.data.mercor.com listing publisher
Assessments and question banks the author or publisher no buyer ask on record n/a
Learner work and forum posts the learner, with a license to the platform none; not course content coursera.org/about/terms
Tutoring transcripts the service terms and student data rules no program on record asks for them n/a

Mercor's listing, published October 17, 2026 and re-read November 17, 2026, asks for "Content the company holds the rights to license for AI training" and names "course creators" among ideal referrers and LMS among its tools.

Mercor states USD 35,000 to 370,000, "Depends on catalog size and rights", as a range for the whole company deal, not per hour, title or course. The referral terms are on Mercor's publisher or education listing profile.

Who Holds the Training Right in a Course: Instructor, Institution or Platform

The right to license a course for AI training sits with whoever owns the copyright in it or has been granted that right in writing, and for an employee's course the Copyright Act makes that the employer unless a signed writing says otherwise (17 U.S.C. 201(b)).

"Training right" is this page's plain term for the permission to use content to train a model, not a statutory term. The table compares four situations by who decides it, as of November 17, 2026.

Situation Who decides the training right Wording read Source
Employee instructor, course made within the job the employer, unless a signed writing says otherwise 17 U.S.C. 201(b); section 101 "work prepared by an employee within the scope of employment" govinfo.gov
Freelance or guest instructor a written assignment or license; commissioned works count as made for hire only in nine listed categories with a signed writing 17 U.S.C. 101 govinfo.gov
Course hosted on a platform the agreement between the platform and the content provider Coursera: such offerings "governed by the relevant agreements in place between Coursera and Content Providers" coursera.org/about/terms
Open courseware the published license MIT OpenCourseWare: CC BY-NC-SA 4.0 with an AI training section ocw.mit.edu

Whether a recorded lecture or a syllabus falls within the scope of employment, and what a faculty IP policy adds, depends on the institution's own policy and contract.

No institution's policy was read for this page and none is described. Ownership of research data inside a university is covered on the page about universities and research groups licensing data, and this page does not restate it.

What a Course Platform and an Open Courseware Site State About AI Training

On November 17, 2026 the two education sites whose terms could be read both state a position on AI training: Coursera bars training AI models on its content without prior written consent, and MIT OpenCourseWare permits AI training only for non-commercial models under its Creative Commons license.

The table compares four sites, alphabetical, by the document read and what it states. "Nothing recorded" means the page could not be read, not that AI use is allowed or barred.

Site Document read, date read What it states on AI training Source
Coursera Terms of Use, effective 2026-01-01, with Acceptable Use Policy, read 2026-11-17 users may not, without prior written consent, use content for text or data mining or to "develop or train" AI and machine learning models "irrespective of the purposes whether commercial or non-commercial"; learners keep rights in their submissions coursera.org/about/terms
Khan Academy terms page returned script-only content, read 2026-11-17 nothing recorded khanacademy.org
MIT OpenCourseWare Privacy and Terms of Use, "Last Updated: 08/11/2026", read 2026-11-17 AI training permitted "unless otherwise specified" with attribution, non-commercial use only and ShareAlike under CC BY-NC-SA 4.0 ocw.mit.edu/pages/privacy-and-terms-of-use
Udemy terms pages returned HTTP 403, read 2026-11-17 nothing recorded udemy.com

Coursera's Acceptable Use Policy binds users of its services. MIT OpenCourseWare's section "Permitted Use of AI Training" requires attribution, states that models trained on its content "shall be used for non-commercial purposes", that "Any commercial application of OCW content through AI training is strictly prohibited", and that derivative models carry the same or a compatible CC license.

It adds that "Determination of commercial vs. non-commercial purpose is based on the use, not the user". The date is quoted as printed, without a day-month reading.

These are the sites' terms for people who use their content, not a license to anyone and not a statement of what a content provider may license under its own agreement.

Mercor's listing names "Content that is already open access" as not a fit, and Creative Commons' position on AI training is set out with Creative Commons and other dataset licenses.

Where Course Content Ends and Student Data Begins

A course license covers what the course owner created, and the assignments, forum posts, recordings of students and tutoring chats around it belong to a different set of rights and rules, which the course owner cannot license by owning the course. The rules for licensing education records for AI are on their own page.

The table compares four kinds of student-side data by the rule that decides, as of November 17, 2026.

Student-side data Rule that decides Wording read Source
Assignments and forum posts the learner's rights and the platform's license Coursera: learner retains rights; Coursera holds a "perpetual, sublicensable, non-exclusive, worldwide license" coursera.org/about/terms
Recordings that show or record students FERPA if an education record; biometric laws for voice and face 34 CFR 99.31(b)(1) de-identified release ecfr.gov
Tutoring transcripts with children COPPA; FERPA where a school is involved AI-training disclosures need separate verifiable parental consent, 90 FR 16950 federalregister.gov
De-identified student-level data 34 CFR 99.31(b)(1) removal of all PII and a reasonable determination of non-identifiability "whether through single or multiple releases" ecfr.gov

Coursera's terms treat homework, quizzes, exams, projects, other assignments and forum posts as User Content, in which learners retain all intellectual property rights, and exclude course content that content providers or their instructors make available.

The learner grants Coursera a "fully-transferable, royalty-free, perpetual, sublicensable, non-exclusive, worldwide" license, so a learner's work is not part of an instructor's course.

The amended COPPA Rule (90 FR 16918, effective June 23, 2025, compliance date April 22, 2026) treats disclosing a child's personal information to train AI as not integral to the service, which requires separate verifiable parental consent.

See COPPA and children's data for the rule, and note that students' voices, faces and handwriting also raise biometric rules.

What Education Publishers Have Put on the Record

The record on education publishers is thin: Wiley states AI licensing revenue of USD 16.0 million in its Learning segment for the fiscal year to April 30, 2026, and no other education publisher's AI licensing terms or values are in the sources read. Wider terms and filings are collected under academic publishers licensing to AI companies.

The table lists what is on record by party, alphabetical, with every figure a claim by its stating party, as of November 17, 2026.

Party What is on record Status or date Source
Cengage Learning and McGraw Hill plaintiffs with Elsevier, Hachette, Macmillan and Scott Turow v. Meta Platforms, N.D. Cal. 3:26-cv-11415-VC; Cengage also plaintiff v. Google, S.D.N.Y. 1:26-cv-05870-LAP Meta suit filed 2026-05-05, pending as of 2026-11-09; Google suit filed 2026-07-10, pending as of 2026-11-17 court dockets
Mercor states USD 35,000 to 370,000, "Depends on catalog size and rights" listing published 2026-10-17, re-read 2026-11-17 api.data.mercor.com
Wiley states Learning AI licensing revenue USD 16.0 million, prior year approximately USD 29 million 10-K, fiscal year to 2026-04-30 sec.gov

Wiley's Research segment and its combined total are on other bases and are kept on the Wiley pages. The Meta suit, filed May 5, 2026 and transferred to N.D. Cal. on September 25, 2026, was pending as of November 9, 2026.

In the Google suit, Google filed a motion to dismiss on October 29, 2026, and as of November 17, 2026 the case is pending. The claims concern copies used in training, and this page does not say which works are textbooks.

Licensing and litigation run side by side, and the case pages hold the statuses. No Pearson, Chegg, Instructure or Khan Academy licensing record is in the sources read, and none is named here as a licensor.

The company angle on edtech and course companies as a sector is on the page about education companies licensing content to AI.

Clearing a Course Catalog Before Anyone Is Asked to License It

Before a course holder offers content to a buyer, it maps for each course who made it, under what contract, and what third-party material is inside.

A rights audit is the general method, and Mercor's own scoping questions ("What do you hold the training rights to?", "How many titles, courses or hours?", "Which subjects and languages?", "Is any of it already licensed for AI training?") show what a buyer asks.

  1. Record the author and contract for each course: employee, contractor or guest speaker.
  2. Read the platform or distributor agreement and note whether it grants the platform exclusive or AI rights.
  3. List third-party material inside slides, readings and video, the case Mercor calls content "distributed under someone else's license", and check it against field of use and scope limits.
  4. Record the people on screen and their releases, with student voices and faces handled under the student data limits above.
  5. Count versions, languages and transcripts.

The result is a rights map, not a conclusion that everything can be licensed.

Licensing Course Content: Benefits, Risks, Downsides and Alternatives, Including Not Licensing

Licensing course content the holder owns can earn from material already produced, but a course often has co-holders and student data inside it, and a trained model keeps what it learned.

Benefits:

  • A named buyer ask exists for course material and lecture video (Mercor listing, re-read November 17, 2026).
  • Mercor states a deal range as a claim, which gives a rough scale for a catalog.
  • Content already produced and structured, with LMS tools and transcripts, fits the listing's tools and inclusions.

Risks:

  • Breach of a platform or distributor agreement, which governs hosted courses.
  • A co-holder's or instructor's unlicensed material inside a course.
  • Student data inside recordings and forums.
  • A CC non-commercial or ShareAlike license on borrowed material.

Downsides:

  • No public price per hour or title exists.
  • Clearing work is needed course by course.
  • A trained model cannot be untrained.

Alternatives:

  • License only fully owned courses.
  • License text and transcripts without learner material.
  • Ask the platform for its written position.
  • Keep the content for in-house products, or decide on whether to sell or license data at all, since not licensing is a valid outcome.

No option removes the need to check the contract first.

Neighboring Content Types, Holders and the Seller Steps for Course Content

This page covers course and education content as a data type; the publishers, universities, companies and laws around it each have their own page. It is information, not legal advice. The overview of data types AI companies buy lists every type.

Neighboring content types and rules:

Seller steps for course content:

  1. Check data sale readiness.
  2. Read the license clauses in AI data licensing agreements.
  3. Put NDAs before sharing data with an AI buyer in place.

Frequently Asked Questions

Who owns the AI training rights to a recorded course?

The author or an assignee owns them, and for an employee's work within the scope of employment the employer does (17 U.S.C. 201(b)). A platform agreement or an institutional policy can change this.

Can MIT OpenCourseWare content train a commercial model?

No: MIT's page states that AI training is allowed only for non-commercial models, with attribution and ShareAlike. It reads "Any commercial application of OCW content through AI training is strictly prohibited".

Are student assignments part of a course license?

No: Coursera's terms leave learners their rights in submissions. Student records have their own rules under FERPA, set out on the education records page.

How much do AI companies pay for course content?

No price per hour or title is on record. Mercor's whole-deal range is its own claim, checked November 17, 2026. Methods are in what company data is worth to AI companies.

Advisers and agents can use the introducer form.