Ask any acquisitions editor what eats their week now. Managing AI-generated submissions will come up fast. A generative tool can spit out a clean, query-ready manuscript before lunch, and the slush pile has changed because of it. More files arrive. They're more polished. And the old trick of skimming for typos and clumsy sentences no longer tells you who the serious writers are. This guide is about handling that without grinding down your readers or rejecting honest authors who happen to run a grammar checker.
I'm not here to argue that technology is the enemy, or that every author is hiding something. The job is responsible triage. You want to know when machine-written text is probably in front of you, treat any detection signal as a hint rather than a ruling, and protect the standards and the original voice your imprint was built to publish.
How AI Is Changing the Slush Pile
A few years back, a flooded inbox was a happy problem. It meant your call for submissions landed. Now a flooded inbox might mean one person ran the same prompt across ten genres and hit send on every one. The cost of submitting has basically collapsed. That cost didn't disappear, though. It moved onto the people reading.
Here's what most editors keep running into:
- More files, less variation. Volume is up, but a surprising share of manuscripts read the same way. Competent. Even-paced. Strangely free of the rough edges that mark a real voice still finding itself.
- Polished surface, hollow middle. Machine drafts rarely have spelling errors or runaway sentences. What they have is worse. Plot logic that won't hold up, characters who never once contradict themselves, and confident "facts" in nonfiction that simply aren't true.
- Template sameness. Synopses and opening chapters keep hitting the same beats in the same order, because they came from the same widely shared structures.
- Spray submissions. A writer who once agonized over a single query now sends thirty, since each extra one costs almost nothing to make.
None of this means AI involvement automatically kills a manuscript. Skilled authors use these tools for brainstorming, line edits, fact-finding, all the time. The problem is the undisclosed, unverified, fully machine-written work that shows up wearing the costume of original craft.
Set Your Policy Before Managing AI-Generated Submissions
The single most useful move in managing AI-generated submissions happens before anyone scans anything. You decide, in writing, what your house actually permits. A clear policy protects you legally, tells authors where they stand, and hands your readers one consistent rule to apply across the pile.
Pick where your imprint sits on the spectrum, then say so plainly on your submissions page:
- Disclosure-based. Authors may use AI but have to disclose how and where. This one's getting common, and it pairs cleanly with originality warranties in your contracts.
- Assistive-only. AI is fine for spelling, grammar, and research support. The prose and the creative work have to be the author's.
- Human-authored only. The manuscript must be substantially written by the named author, with AI use limited or off the table.
Whatever you land on, put it in three spots. The public guidelines. The submission form, ideally a checkbox that forces an explicit disclosure. And the contract, as an originality and AI-use representation. A writer who knowingly ticks "this is my original work" has handed you firm ground to act on if that claim falls apart later. That's a far stronger position than a detector score by itself.
A Practical Triage Workflow for Editors
Detection should make your readers quicker. It shouldn't turn them into forensic investigators. You want human attention spent where it pays off, on the promising stuff, and you want the obvious machine submissions cleared fast and fairly. Here's a flow you can repeat.
Step 1. Read first, scan second. Your editorial gut is still the best instrument in the building. A manuscript that feels flat, generic, or oddly sure of itself about everything has earned a second look. Let that instinct pick what to scan, not the reverse.
Step 2. Run the full file through a detector. When something feels off, push the manuscript through a Document Detector for an origin signal on the text as a whole. It returns an overall probability and can flag where the writing reads as machine-generated. Use that to aim your reading. Never as a standalone reason to reject.
Step 3. Check originality, not only AI likelihood. A piece can be human-written and still plagiarized. It can be AI-assisted and still genuinely the author's. Run anything questionable through a Plagiarism Checker to catch lifted passages, recycled public-domain text, and content scraped from elsewhere. That's a separate risk, and AI scanning alone won't surface it.
Step 4. Pressure-test the substance. For fiction, hunt for internal consistency. Do characters keep stable motivations? Does the timeline hold? Does the ending earn its setup? For nonfiction, verify a sample of claims, quotes, and citations against real sources. Confident fabrication is the signature failure of machine-written text, so this step catches what a polished surface hides.
Step 5. Match signals to your policy. Pull it together. The detection signal, the originality result, the substance check, the author's disclosure. A clean disclosure plus solid craft is a green light, even if AI helped. A "fully original" claim that fights with strong detection and plagiarism signals is your cue to investigate or pass.
Step 6. Log the decision. Keep a short note of what you scanned and what you found. Consistency across readers is exactly what makes your process defensible and fair to every author who submits.
Reading the Signals Honestly
Detection tools earn their keep because they're fast. But they're probabilistic, and treating a score as proof is how good imprints make unfair calls. A handful of principles keep your judgment honest.
AI detection estimates likelihood by reading patterns in word choice and sentence rhythm. That works well at scale. It also has real limits. False positives happen, and they tend to cluster around the writers you least want to lose. Non-native English speakers, whose careful, measured prose can read as "machine-like." Debut authors leaning hard on grammar tools. Genre writers working inside tight conventions. Know where detection is weakest and you'll read its output more fairly. Our note on AI detection limitations is worth passing around your reading team.
The discipline that actually protects authors is simple. A detection score is a reason to read more carefully. It is never a verdict on its own. No serious editor rejects a manuscript because a tool returned a number. Combine the signal with your own reading, the originality check, and the author's disclosure, and only then decide.
Protecting Voice, Authors, and Your Imprint
Strip away the mechanics and managing AI-generated submissions comes down to defending the one thing that makes a publishing house worth submitting to. A reputation for genuine, original work.
- Reward disclosure. Don't punish it. An author honest about light AI help should never fare worse than someone who used more and stayed quiet. Make transparency the safe choice, every time.
- Protect the writers you'd hate to lose. Build false-positive awareness into reader training, so careful or non-native prose doesn't get mistaken for machine output. Wrongly rejecting a real talent costs you a lot, and you never see the bill.
- Verify before you contract, not after. Catching an originality or authorship problem in triage is cheap. Catching it after publication, through a reader, a reviewer, or a rights dispute, is expensive and it follows your name around.
- Mind confidentiality. Unpublished manuscripts are sensitive author property. Before you run any submission through a third-party tool, confirm how that provider handles data, and favor vendors who are transparent about how they store and use uploaded files.
- Keep the human in charge. Tools triage. Editors decide. The value of your imprint is human taste applied consistently, so let detection clear the noise and free your readers to spend their attention on the signal.
Frequently Asked Questions
Can a detector prove an author used AI to write their book?
No. AI detection returns a probability, not proof. Careful, formulaic, or non-native prose can read as machine-generated and trip a false positive. Use a detector to decide where to focus your reading, then lean on substance checks, originality results, and the author's own disclosure before you reach any conclusion.
Should we ban AI from submissions entirely?
Most houses find a disclosure-based or assistive-only policy more workable than a flat ban, because skilled authors legitimately use AI for editing and research. Whatever you pick, state it clearly in your guidelines, capture an explicit disclosure on the submission form, and back it with an originality warranty in the contract.
What's the difference between AI detection and a plagiarism check?
AI detection estimates whether text was likely machine-generated. A plagiarism check finds passages copied from existing sources. They catch different problems, so reading teams should run both. A manuscript can be human-written but plagiarized, or AI-assisted but genuinely the author's own.
How do we handle a high-detection score from a promising author?
Treat it as a prompt to look closer, not a rejection. Re-read for voice and substance, run an originality check, and revisit the author's disclosure. If the craft is strong and the disclosure is honest, the score alone isn't a reason to pass. If the signals clash with a "fully original" claim, ask the author directly.
Make Triage Faster Without Lowering Your Standards
AI is part of submissions now, and that's not inherently a threat. Undisclosed, unverified machine-written work is. Set a clear policy. Read before you scan. Treat every detection score as probabilistic guidance. Check originality alongside AI likelihood. Do those four things and your slush pile keeps moving while your imprint's standards stay intact.
Streamline your slush pile by running manuscripts through the Document Detector and confirming originality with the Plagiarism Checker before any submission reaches a contract.
Try it on your own writing