Should Authors Use AI?

It depends entirely on what you use it for. Using AI to brainstorm, research or format carries little consequence. Using it to generate the prose you publish creates three concrete problems: purely AI-generated text isn't protected by copyright in the US, retailers including Amazon require you to disclose it, and readers increasingly treat it as a defect. The question isn't ethical in the abstract — it's which specific task, and what you'll have to declare.

The longer version

Treating "AI" as one decision is why this argument goes nowhere. Sorted by what actually happens to you:

Low consequence

Brainstorming and rubber-ducking a plot problem
Research starting points you then verify independently
Reformatting, converting files, cleaning up a table
Summarizing something you wrote to draft a synopsis

Nothing here ends up in the published text as machine-written prose. The main exposure is confidentiality — see below.

Real consequence

Generating prose that ships in the book
Generating your cover art
Generating your blurb or ad copy
Generating reviews or reader-facing content

Three specific problems, in order of how concrete they are:

1. Copyright. The US Copyright Office's position is that copyright protects works of human authorship. Material generated by AI without sufficient human authorship is not protectable. A work containing both remains registrable — but you must disclaim the AI-generated portions in the application. Practically: the parts a machine wrote are not yours to defend, and misrepresenting this in a registration is its own problem.

2. Platform disclosure. Amazon KDP requires publishers to disclose AI-generated content when publishing. It distinguishes this from AI-assisted content — where you created the work and used AI to refine it — which does not require disclosure. Read the current definitions rather than trusting a summary, including this one; the line between the two categories is exactly where the ambiguity lives.

3. Reader reception. This one isn't regulated and matters most. In fiction especially, the disclosure that a book was machine-written is treated by a large share of readers as grounds not to buy, and by reviewers as grounds to say so publicly. Whether that's fair isn't the question if you're deciding how to spend the next year.

The exposure everyone forgets

Separate from whether AI writes anything: pasting your manuscript into a tool sends it to that tool's servers. Whether it can then be used for training depends on the terms and the account tier, and the answer differs between consumer and business plans of the same product.

This applies even to the low-consequence uses. Asking a chatbot to summarize your chapter means your chapter has left your machine. If the work is unpublished, that's the decision to be deliberate about — more than the ethical argument, which at least you can see coming.

See How to Keep Your Manuscript Private in the Age of AI.

The context authors are deciding inside

Two survey findings, because they explain why this is charged rather than technical:

Authors Guild, December 2023: 96% of 2,400+ surveyed writers said consent should be required, and authors paid, if their work is used to build an AI system.
University of Cambridge, Minderoo Centre, 2025: 59% of surveyed published novelists believed their work had already been used to train large language models without permission, and 93% said they would opt out if given the choice.

That doesn't settle what any individual author should do. It does explain why the reader reaction in point 3 is as strong as it is.

Common exceptions

Accessibility uses — text-to-speech, dictation, screen readers — are not what this debate is about, and treating them as equivalent is a mistake.
Translation is genuinely contested. Machine translation with professional human editing is standard in some markets and disclosed in others.
Non-fiction and technical writing face weaker reader objection than fiction, though the copyright and disclosure rules are identical.
Work-for-hire and traditional contracts increasingly include explicit AI clauses. Check before signing, and check whether the clause covers assistance or only generation.
On-device tools — local spellcheck, on-device text-to-speech — send nothing anywhere and are a different category from cloud services, whatever they're branded as.

Sources

US Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (2023), and Copyright and Artificial Intelligence, Part 2: Copyrightability (2025).
Amazon KDP Help, AI-generated and AI-assisted content disclosure requirements.
Authors Guild survey, December 2023; University of Cambridge Minderoo Centre survey, 2025.

General information, not legal advice. Platform policies change frequently — verify against the current KDP help pages before publishing. Verified 9 August 2026.