Do Publishers Accept AI-Assisted Books? What the Writer Should Know Before Submitting
Do publishers accept AI-assisted books? In 2026 the question stopped being theoretical: a major publisher cancelled a novel over suspected AI use, and a number of literary agencies amended their submission guidelines. In this article, we take a look at where publishers and agents draw the line, what your contract asks you to promise, and when the writer should disclose AI use.
There's nothing new about writers using tools. Spellcheckers, grammar checkers, and word processors have been part of manuscript preparation for decades, and writers are not expected to disclose them. That being said, generative AI is a different kind of tool—it can produce the text itself. Publishers, agents, and contracts now turn on that difference—between a tool that helps the writer and a tool that writes.
Remember, the most important considerations are what the agent, the publisher, and the contract require. The general picture, however, is now clear enough to guide the writer before any submission goes out.
AI-assisted and AI-generated: the distinction publishers draw
Two terms carry most of the weight in this discussion. AI-generated text is text that an AI tool produced. AI-assisted work is work the writer produced, with an AI tool in a supporting role: checking spelling, flagging a weak scene, or suggesting where a chapter loses momentum.
Amazon's Kindle Direct Publishing guidelines give one of the clearest public definitions. KDP treats content as "AI-generated" if an AI-based tool created the actual text, "even if you applied substantial edits afterwards." It treats content as "AI-assisted" if the writer created it and used AI tools to "edit, refine, error-check, or otherwise improve that content," or used AI to brainstorm while writing the text themselves. KDP requires writers to disclose the first and not the second.
Traditional publishing is usually stricter. The Authors Guild advises authors not to use AI to write for them at all: "If you are claiming to be the author of a work, the words should be your own." Several literary agencies go further still, as we will see below. The writer should therefore treat KDP's definition as a useful vocabulary, not as a standard every agent and publisher shares.
Do publishers accept AI-assisted books?
For AI-assisted books that involve only the most limited kinds of assistance, the answer is generally yes. Spelling and grammar checks and research are rarely the concern, and feedback that the writer acts on in their own words is usually treated as acceptable, although some agencies exclude AI editing tools as well. Generated prose is another matter, and the consequences of submitting it without disclosure can be severe.
The warranty of originality in your contract
Standard publishing contracts include representations and warranties: legally binding promises the author makes about the book. One of them is that the work is original. As the Authors Guild explained in April 2026, "original" carries its copyright meaning here—created by a human author. The Guild warns that if a book contains "more than a minimal amount of AI-generated text" that was not disclosed, "the publisher may have the right to terminate the entire agreement on breach-of-contract grounds," and that "an entirely AI-generated plotline or wholesale adoption of AI-generated characters may violate the warranty."
Its advice is direct: disclose to your publisher any AI-generated text, characters, or plot in the manuscript, and make sure any agreed AI use is reflected in the contract. The Guild writes from a US legal perspective, and contracts differ, so an agent or a lawyer should read the specific clause before you sign.
What the Shy Girl cancellation changed
In March 2026, Hachette cancelled the US publication of Mia Ballard's horror novel Shy Girl, due from its Orbit imprint, and discontinued sales in the UK, after allegations that the novel had been written with generative AI. Ballard told The New York Times that she had not used AI to write the novel, but that someone she had hired to edit it had.
The case showed the writer two things. First, the consequences of suspected AI use can arrive after a book has been acquired, edited, and published in one market. Second, under the warranty, the author is usually the one answerable for the text, whoever introduced the generated material. An editor, a friend, or a paid service that runs your prose through a rewriting tool has put AI-generated text into your manuscript, and the warranty you sign is still yours.
Detection, accusation, and the burden on the author
Detection is imperfect, even where a detector's false-positive rate is low. In July 2026, The Atlantic reported on a study that ran thousands of Kindle e-books through the AI detector Pangram and singled out H. M. Wolfe's Daggermouth, which had been self-published and then acquired by Simon & Schuster's Scarlett Press in a seven-figure, two-book deal. The study's figure combined text the detector labelled AI-generated with text it labelled "moderately AI-assisted," and the researchers acknowledged that such numbers can tell only so much. Wolfe denies using AI; in a statement quoted by The Atlantic, she called the suggestion "wholly untrue" and said, "I wrote this book myself." Simon & Schuster stood behind its author.
In August 2026, by contrast, Publishers Weekly reported that a novel was pulled from submission by the author's agents over concerns about AI use, after an auction and an accepted offer from Minotaur. The author denies the accusation.
The burden in both cases fell on the author. An accusation requires only a detector score; a defence requires evidence of how the book was written. Keep your drafts, your notes, and your version history. They are the most practical protection a writer has.
Why literary agents are banning AI in submissions
Agents read submissions for voice, and voice is what generated text tends to flatten. In February 2026, The Bookseller reported that a number of agencies were amending their guidelines to urge writers not to use AI. Greene & Heaton's submissions page now states that submissions "originated, written or edited using Artificial Intelligence (AI) will not be accepted," and applies that rule to the cover letter, synopsis, and proposal as well as the manuscript. The page adds that "even what might seem, to you, to be harmless and helpful AI editing tools can really flatten your writing."
Other agents use different wording. Annabel Barker asks writers who used AI "for any aspect" of a submission to say so in the query, and does not accept submissions that used AI in the written or illustrative work. Michelle Witte does not accept "any submissions that used AI in its creation." Paige Terlip states plainly that she does not accept AI-generated submissions for art or text.
The lesson for the writer is that the agent's guideline, not a general industry rule, decides the question. Generated queries are also part of the volume that has led many agents to close their inboxes. Where an agency excludes AI editing tools, read that as excluding any tool that changes your text, and ask before submitting if the wording is unclear.
Feedback versus generated text: where the line usually sits
Read together, the guidance above sorts most AI use into two groups. The test is simple to state: did the tool produce words that now appear in the manuscript?
Uses usually treated as acceptable
Spelling checks, research, and feedback belong here. A feedback tool points to a problem and leaves the writer to solve it. A note that a scene's tension drops in its second half, or that a paragraph repeats what the previous one established, changes nothing on the page until the writer revises in their own words. That is the logic behind paragraph-level critique, and it is close to what a grammar checker does at the level of the sentence. The Authors Guild describes non-text-generating uses such as research and outlining as "generally permissible," while cautioning writers not to copy AI output into the work.
There are exceptions, of course. A guideline that refuses any submission "that used AI in its creation" may be read to include feedback tools, and the writer should respect the strictest reasonable reading of the agent's words.
Uses that require disclosure or are refused
Any text an AI tool produced belongs in this group: drafted chapters, generated dialogue, a rewritten passage, or an AI-written query letter or synopsis. AI-generated plot or characters belong here too, under the Authors Guild's reading of the warranty.
This includes the rewrite features in tools made for writers: a rewrite feature produces AI-generated text, so a passage it rewrites should be disclosed or recast in your own words before it goes to an agent or publisher. Recasting means writing the passage afresh, not lightly editing the tool's version, since text that began as AI output remains AI-generated even after substantial edits under KDP's definition. A rewrite can still show what a fix might look like; the version you submit should be yours.
Is it acceptable to use AI to edit your book?
It depends on what "edit" means. If the tool corrects spelling or points out problems, most publishers are unlikely to object, although some agencies exclude AI editing tools altogether. If the tool rewrites your sentences, the result is AI-generated text and should be treated as such.
Can literary agents and publishers tell if you use AI?
Most of the time, yes. Editors at publishing houses and literary agents read an incredible amount of text. Their livelihoods depend on being able to identify originality and patterns. When it comes to prose, your average literary professional is as well equipped as any to identify AI-generated text. The quality of AI prose will probably improve in the years to come, but none of the major models have thus far been able to shake the "flatness" inherent to their creative-writing output.
Additionally, it is not advisable for a querier to assume they will get away with submitting AI-generated text because there isn't AI-identification software that is 100% correct. Don't think that because a literary professional can't conclusively prove you've used AI that that will stop her from rejecting your proposal—a strong suspicion is more than enough for her to move on. Michelle Witte, for example, notes in her guidelines that she will never run submissions through AI to check for AI use, since such tools "aren't accurate, anyway."
However, as the Shy Girl debacle proves, there are exceptions. Even the largest publishers concede that it can be difficult to identify AI-generated text. Asked at a Publishers Weekly event how publishers could prevent another Shy Girl, Hachette Book Group CEO David Shelley said that "anyone who thinks they know the answers here is lying or delusional, because we're all trying to figure this out."
Even so, the absence of a reliable AI-detection tool is not a reason to take the risk. Agents and editors read closely. For example, UK-based Greene & Heaton's website notes that in many cases the agency is "able to discern widespread use of AI tools." A manuscript that reads as flattened or generic is likely to be declined whether or not anyone can prove how it was made.
If a litereary agent reads ten submissions in one sitting, and seven have an extremely similiar style, is it really difficult to believe that she won't be able to tell the difference?
How to disclose AI use, if you need to
Disclosure works best when it is early, specific, and plain. If an agent's guidelines ask about AI, answer in the query or on the form. If a publisher makes an offer, raise any AI use before the contract is signed, so that the agreed use can be written into it.
Let's have a look at two ways of disclosing the same thing.
In the first example, the disclosure is vague:
I used some AI tools during the writing process.
In the second example, it is specific:
I used an AI feedback tool to identify pacing and structural problems in the draft. All of the text in the manuscript is my own, and no AI-generated text, characters, or plot appear in it.
The first version raises more questions than it answers, and it leaves the agent to imagine the worst. The second names the tool's role, the extent of its use, and what it did not do. If a passage did begin as AI output, say which passage and how it was handled, or—better still—rewrite it yourself before submitting.
Keeping the words your own
Publishers and agents are not, for the most part, objecting to AI-assisted books in which the writer sought feedback. They are objecting to text the writer did not write, submitted under a warranty that says otherwise. The cost of blurring that line can be a declined submission, a cancelled contract, or an accusation that is difficult to answer. The safest course is not to use AI at all. If you do use AI, make sure you don't present any of its text output as your own.
If you want a human editor to review your manuscript before you submit—someone who will help you keep every sentence yours—have a look at Manuscript Mentoring's Book Editing service.
Comments