Can a company license its books and long-form text to AI companies?
Yes, when the company holds the rights to the text. Long documents teach models to keep track of an argument over many pages, which short web text does not. Manuals, technical reports, training courses, books and years of newsletters can all qualify. The question for each title is who owns the copyright and what the author contracts say.
Long-form text a company may own
| Technical manuals and guides | Installation and repair manuals, written in house over many product versions. |
| Research and white papers | Long reports written by staff, including drafts and reviewer comments. |
| Course material | Lessons, workbooks and assessments built for customer or staff training. |
| Books the company published | Titles where the company holds the rights, or where the author contract allows this use. |
| Newsletter and article archives | Years of issues written by staff, with the edits made before each went out. |
Checking the rights title by title
Text written by employees as part of their job usually belongs to the company. Text written by freelancers, outside authors or co-authors belongs to whoever the contract says. Older publishing contracts rarely mention AI training, and many grant the publisher only print or electronic rights.
A rights list helps: one row per title with the author, the contract, the date and what rights it grants. Titles with unclear rights are set aside until the author or estate agrees in writing.
Quoted material inside a work, such as images, excerpts or charts from someone else, may need its own permission.
Why long text is wanted
Models need text that holds an idea across chapters: a manual that defines a term on page three and uses it on page two hundred, a course that builds one lesson on the last. Edited, accurate and complete text is also scarce compared with the volume of web pages.
Drafts and editor comments add another layer. They show how a piece of writing was improved, and few sources have them.
Preparing a set
Native files are better than scans. Word documents, the publishing system files or clean text exports keep headings, tables and footnotes in place. Scanned pages need text recognition and checking first.
Count words or pages per title, years covered and the share where the rights are confirmed. Those numbers start the conversation.
Text that is kept back
| Titles where the rights are unclear | Set aside until the author, estate or co-publisher agrees in writing. |
| Licensed or quoted third-party material | Excerpts, images and charts owned by others are removed or cleared. |
| Client deliverables | Reports written for a paying client usually belong to that client or carry confidentiality terms. |
| Text with personal stories or data | Case studies naming real people are cleaned or left out. |
More on this: how personal details are removed, who carries the risk and how you get paid.
An example, invented for illustration
A small industrial training company with thirty-five staff at its busiest has written fourteen years of course workbooks and four technical books. Staff wrote the workbooks, so the company owns them. Two of the books were written by outside authors under contracts that grant print rights only, so they wait until the authors agree. The workbooks go out as native files with student names removed from the example exercises.
Questions owners ask about this
Does an old publishing contract cover AI training?
Often it says nothing about it, and silence is not permission. Counsel reads the grant of rights; where it is unclear, the author is asked.
Is text already sold to the public still licensable?
It can be. Selling a book gives readers a copy, not the right to train on it, so the rights holder can still license training use.
Should we send scans or the original files?
Original files where you have them. They keep the structure, need no text recognition and leave fewer errors for a buyer to find.
Didn't find your question? Ask us directly.
Keep reading
Related kinds of records
- SOPs and knowledge bases: Procedures with their edit history and comments; published help articles add nothing.
- Theses and dissertations: Authors, universities and sponsors: who can license a thesis, and what to settle first.
Industries that keep these records
- Publishers: Drafts, copyedits and proof corrections for the same passage, limited to titles where the publisher can show it holds the rights.