AI Authorship and Copyright Ownership: Who Owns an AI-Assisted Book?

An increasing number of published books are, to some degree, AI-assisted. An author might have used a large language model to draft an outline, generate alternate phrasings for a stuck paragraph, produce sample dialogue, or clean up sentence structure during revision. That reality raises a question publishers, agents, and authors are all trying to answer with more precision than the law currently offers: when a book is substantially assisted by AI, who actually owns the copyright in it?

The framework that has emerged so far, as of this writing, rests on a few fairly clear principles combined with a great deal of case-by-case uncertainty. This article walks through the ownership analysis, how mixed human/AI works are generally treated, and the disclosure questions authors increasingly need to navigate with publishers and retailers. None of this should be read as a substitute for tailored legal advice — an attorney who reviews your actual manuscript and contract will be far better positioned to advise on your specific situation than any general overview.

US copyright law has consistently required a human author, and the Copyright Office reaffirmed this explicitly with respect to generative AI in guidance issued in 2023. Under that guidance, content generated entirely by an AI system, without sufficient human creative input, cannot be copyrighted by anyone — not the AI company, not the person who typed the prompt, and not a publisher who later distributes the work. There is simply no copyright to own in that specific machine-generated expression.

This means the ownership question only becomes interesting once a human has contributed genuine creative authorship to a work. At that point, ownership follows the same basic rules that have always applied to human-authored works: the author (or the author’s employer, in a work-for-hire arrangement) owns the copyright unless and until it is transferred, licensed, or assigned by contract, such as through a traditional publishing agreement.

How Mixed Human/AI Works Are Generally Treated

Most real-world AI-assisted books aren’t purely one thing or the other — they’re a blend, with some passages substantially human-authored and others closer to raw AI output that received light or no editing. The Copyright Office’s approach to these mixed works is to evaluate authorship at the level of specific expression, not the work as a whole.

In practice, this produces a layered result. An author can generally claim copyright in the portions of a work reflecting their own creative expression — original plot structure they devised, prose they wrote or substantially rewrote, dialogue they crafted — while the specific AI-generated passages that were not meaningfully altered by human creative judgment may fall outside the scope of that protection. The overall book can still be registered and sold, and the protectable elements are still enforceable against copying, but a purely AI-generated passage embedded in an otherwise human-authored book is not, by itself, protectable expression.

This layered approach has a few practical consequences worth understanding:

Selection and arrangement can itself be protectable. Even where individual elements originated from AI suggestions, an author’s creative choices about which elements to use, how to sequence them, and how they interact with surrounding human-authored material can constitute a form of authorship separate from the underlying content itself.

Editing intensity matters more than tool use. The fact that an author used an AI tool somewhere in the process is not, by itself, disqualifying. What matters is whether the final expression in the published work reflects substantial human creative judgment, as opposed to being essentially the AI’s unedited output.

There is no reliable shortcut to full protection. Some authors have hoped that a quick editing pass over AI-generated text would be enough to convert it into fully protectable, human-authored expression. The Copyright Office’s case-by-case, qualitative approach means this is not a guaranteed outcome, and courts have not yet definitively resolved exactly how much revision is required in every context.

Authors registering a copyright for a work that includes AI-generated content are expected to disclose that fact and identify the human-authored portions on the application. This isn’t merely a formality — a registration obtained without disclosing known, significant AI-generated content risks being challenged later if that omission surfaces, since the certificate itself represents a claim about human authorship. Authors working with agents or publishers on registration should make sure whoever files the application has accurate information about how AI tools were used during drafting.

Disclosure in Publishing Contracts

Separately from Copyright Office requirements, publishing contracts themselves have started to catch up with the reality of AI-assisted writing. It is increasingly common to see publishing agreements include representations and warranties in which the author affirms the extent (or absence) of AI involvement in the manuscript, sometimes paired with indemnification language protecting the publisher if that representation turns out to be inaccurate.

Publishers have practical reasons to want this information clearly documented. A publisher’s ability to enforce copyright against infringers, license subsidiary rights, or sell foreign translation rights can all be complicated if there’s uncertainty about how much of the underlying work is actually protectable. An author who is vague or inaccurate about AI use in these representations can create liability exposure for both themselves and their publisher down the line.

Disclosure to Retailers and Platforms

Beyond formal registration and publishing contracts, some retail and distribution platforms have begun implementing their own AI-content disclosure requirements, particularly for self-published and independently distributed work. These policies vary by platform and continue to evolve, generally requiring authors to indicate whether and how AI tools were used in generating text, images, or other elements of a published work. Authors distributing through multiple platforms should check each platform’s current policy directly, since requirements differ and are subject to change with little notice as platforms respond to both legal developments and marketplace pressure from readers and other authors.

Practical Takeaways for Authors

A few habits can help authors protect their position on ownership questions:

Document your process. Keeping records of drafts, prompts, and revision history creates evidence of the extent of your creative contribution, which can matter for registration, contract representations, and potential disputes.

Be accurate in every disclosure context. What you tell the Copyright Office, your publisher, and any retail platform about AI use should be consistent and truthful. Inconsistent statements across these contexts can create their own legal problems independent of the underlying authorship question.

Understand what you’re actually assigning. When you sign a publishing contract, understand which rights you’re transferring and how any AI-related representations in that contract could affect your liability if authorship of part of the work is later challenged.

Revise substantively, not superficially. If protecting the copyright in AI-assisted passages matters to you, treat AI output as a first draft that requires meaningful creative rework, not a finished product needing only cosmetic changes.

Frequently Asked Questions

You own the copyright in the portions reflecting your own human creative authorship. Passages generated by AI without substantial human creative input generally are not protectable by anyone, meaning the overall work’s protection may be limited to the human-authored elements rather than covering every sentence equally.

Does my publisher own the AI-generated parts of my manuscript instead of me?

No one owns unprotectable AI-generated expression outright, including the publisher. Publishers typically acquire rights only in copyrightable material through the publishing agreement, so purely AI-generated passages that lack sufficient human authorship generally remain outside anyone’s exclusive ownership.

Do I need to tell my publisher I used AI while writing?

Increasingly, yes. Many publishing contracts now include specific representations about AI use, and providing accurate information helps protect both you and your publisher from later disputes about the work’s authorship or enforceability.

Can a retailer reject my book if I don’t disclose AI use?

Possibly, depending on the platform. Some retail and distribution platforms have implemented AI-content disclosure policies with their own enforcement mechanisms. Because these policies vary and change frequently, authors should review the current terms of each platform they use before publishing.

How much do I need to edit AI-generated text before it counts as my own work?

There is no official percentage or fixed threshold. The Copyright Office looks at whether your creative judgment shaped the final expression, not how many words were technically changed, so substantive creative revision matters more than surface-level editing.

Keep detailed records of your drafting process and consult a licensed intellectual property attorney before finalizing registration or signing a publishing contract. This is a genuinely unsettled area, and individualized legal advice is far more reliable than general guidance for a specific manuscript.

This article is provided for general informational purposes and does not constitute legal advice. Authors should consult a licensed attorney regarding the copyright status of their own AI-assisted work. For the underlying statutory framework, see the U.S. Copyright Office’s Title 17 resource.