How to Format an AI-Assisted Manuscript for Submission

The invisible mess hiding in your manuscript

Your manuscript reads clean. You've done your passes, the prose sings, the plot holds together. Then an agent opens the file and within three pages hits a curly apostrophe sitting next to a straight one, an em-dash that's actually three hyphens, and a scene break that's sometimes "***" and sometimes just a blank line. None of this touches your writing quality. All of it screams unedited before anyone reads a single sentence of dialogue.

This is the specific tax AI drafting tools charge you, and almost nobody talks about it. When you generate chapters across multiple sessions, switch models partway through, or paste content between a chat window and your manuscript file, you're stitching together text that was never formatted by one consistent hand. Spellcheck won't catch it because nothing is misspelled. Your own eyes won't catch it because reading for story and reading for formatting are different cognitive modes, and you've been doing the first one for months. The inconsistency is invisible until it's the first thing a professional reader sees.

A manuscript with perfect prose and sloppy formatting reads, to an editor, like a brilliant pitch delivered with spinach in your teeth. The content isn't the problem. The distraction is.

This guide is about the pass almost everyone skips: the mechanical, unglamorous sweep for artifacts that AI output leaves behind. It's quick once you know what you're hunting for. It's also the single highest-leverage hour you can spend before a submission, because it's the difference between a manuscript that reads as professional and one that reads as a draft someone forgot to finish.

The five artifacts that give away an AI-assisted draft

Every AI model has formatting habits, and those habits drift depending on which tool generated which section. Here's what to actually look for.

1. Markdown leftovers

Asterisks for emphasis (*italics* or **bold**) that never got converted to real formatting. Pound signs before chapter headings. Bullet dashes that snuck into narrative paragraphs because the model was thinking in list structure for a beat. These are the easiest to spot and the easiest to miss, because your brain auto-corrects "*gasped*" into emphasis even when the document shows raw asterisks.

2. Mixed dash styles

This is the big one. AI models slide between em-dashes (—), en-dashes (–), and double hyphens (--) depending on context window, model version, and sometimes just randomness. You'll have interrupted dialogue punctuated with an em-dash on page 12 and the same construction punctuated with a hyphen on page 340. Readers won't consciously register this, but editors trained to spot inconsistency will catch it in seconds.

3. Quote-mark drift

Curly quotes (" ") versus straight quotes (" "). This happens constantly when you paste text from a chat interface into a word processor, or when you bounce a chapter through a plain-text editor for a revision pass. One paragraph has proper typographic quotes, the next has straight ones, and somewhere in the middle there's a single apostrophe in a contraction that didn't match either.

4. Inconsistent scene breaks

Some chapters use ***, others use #, others use a plain blank line, others use "~". If you generated scenes across different sessions or stitched together outputs from different prompts, your scene-break convention probably isn't a convention at all — it's a record of your process.

5. Trailing whitespace and phantom line breaks

Extra spaces at the end of lines, double paragraph breaks where there should be one, paragraph breaks missing where dialogue changes speaker. These don't show up visually in most word processors until you convert the file, at which point they turn into ugly, unpredictable spacing in the final output.

None of these are writing problems. All of them are what separates a manuscript that feels professionally prepared from one that feels like a transcript. If you want the deeper context on why AI drafts develop these particular tics, the beginner handbook for writing fiction with AI covers the underlying generation quirks that cause them.

Prompting AI to audit its own mess

Here's the part people miss: the same AI that created the inconsistency can find it, if you ask the right way. Don't ask it to "fix formatting" — that's vague enough that it'll rewrite your prose while it's in there. Ask it to audit and report, not edit. You want a diagnostic pass before you authorize any changes.

This works because it forces the model into a close-reading mode instead of a generative one. You're not asking it to "clean this up" — you're asking it to produce an inventory, which you then act on deliberately. It also surfaces patterns you wouldn't catch manually, like discovering that your model defaults to en-dashes for interruptions in dialogue but em-dashes for asides in narration, which is exactly the kind of inconsistency that's invisible until named.

Once you have the inventory, run a second, narrower pass that actually executes the fix — but keep it mechanical, not creative:

The "flag anything ambiguous" instruction matters more than it looks. It stops the model from silently making a judgment call on a weird edge case — like a dash that might be a typo for a hyphenated compound word rather than an interrupted sentence — and instead puts that decision back in your hands. This is the same principle behind a good five-pass revision order for AI-assisted novels: each pass has one job, and you don't let a mechanical pass quietly become a creative one.

If you're working inside Entangled Text rather than bouncing between a chat window and a separate document, this audit-then-execute workflow is tighter because the model has the full manuscript context in one place rather than fragments you've pasted around. Worth comparing how that context difference plays out if you're curious — see Entangled Text vs ChatGPT for the specifics.

Building your own find-and-replace checklist

AI auditing gets you 90% of the way. The last 10% is house style, and house style is yours to define, not the model's. Different genres and different markets have different conventions, and "correct" formatting means something different depending on where the manuscript is headed.

Before you run a final find-and-replace pass, decide these things explicitly:

• Dash style for interruptions vs. asides. US convention typically uses an unspaced em-dash for both. UK convention often uses a spaced en-dash instead. Pick one and apply it everywhere — don't let your manuscript use US style in dialogue and UK style in narration because two different chapters came from two different sessions.
• Quote style. US fiction almost always uses double quotes for dialogue with single quotes for nested quotes inside. UK fiction frequently reverses this. If you're submitting to a UK agent or publisher, this one detail signals whether you've actually researched their market.
• Scene break symbol. Pick one marker and one spacing convention (centered, with blank lines above and below) and apply it manuscript-wide. If you're not sure what's standard for your genre, check a handful of recently published comps on your shelf — not older books, since conventions shift.
• Numbers, titles, and ellipses. Spelled out or numeral? Three periods or the single ellipsis character (…)? These are small, but inconsistency in small things is exactly what a sloppy-draft radar picks up on.
• Trailing whitespace and spacing after periods. Single space after periods is standard now. Double spaces are a leftover habit some models pick up from older training text, and they're invisible until the manuscript gets reflowed into a different format.

Once you've decided, write the list down as an actual document — a one-page house style sheet — and keep it next to your story bible that AI models actually follow. That story bible already keeps your characters and world consistent; your style sheet does the same job for mechanics. Both exist so you're not relying on memory across a 90,000-word draft.

Here's a prompt that turns your style sheet into an enforceable pass:

Notice this prompt treats style as data, not taste. You're not asking the AI to guess what sounds right — you're handing it a specification. That's the same discipline that makes genre-specific tools like the romance novel with AI assistant or the LitRPG writing assistant work well: specificity in, consistency out. The vaguer your instruction, the more the model falls back on its own training defaults, which is exactly the drift that caused the problem in the first place.

Genre conventions you can't skip

Formatting isn't fully genre-neutral, and this is where a generic style guide fails you. A few examples worth knowing before you finalize your checklist:

• Thrillers and mysteries often use short, isolated paragraphs and scene breaks mid-chapter far more aggressively than literary fiction — if you're working with an thriller with AI workflow, check that your scene-break density actually matches the pacing convention of the subgenre, not just the mechanical marker.
• LitRPG and progression fantasy frequently include stat blocks, system messages, or interface text that's supposed to look distinct from prose — these need their own deliberate formatting rule, not accidental markdown you forgot to clean up.
• Historical fiction sometimes uses period-appropriate spelling or punctuation conventions deliberately. Make sure your find-and-replace pass doesn't "correct" an intentional archaism back to modern standard — this is a real risk if you're leaning on tools built for historical fiction with AI and running a blanket cleanup pass afterward.
• YA manuscripts tend toward shorter paragraphs and more frequent white space breaks for pacing — if you're writing a YA novel with AI, don't let a formatting pass flatten that rhythm into denser paragraph blocks because the model defaulted to "standard" structure.

The point isn't that every genre needs wildly different mechanics. It's that "fix the formatting" is never actually a neutral instruction — it always implies a convention, and you need to be the one who sets it rather than letting the model assume.

The final scan: what conversion itself breaks

Here's the twist almost nobody accounts for: even after your manuscript is clean, the act of converting it into a submission-ready format can reintroduce the exact problems you just fixed. Pasting from a word processor into a plain-text submission portal can silently convert curly quotes back to straight ones. Exporting to EPUB or MOBI can collapse your carefully centered scene breaks into left-aligned text. Converting a docx to PDF can shift trailing whitespace into visible gaps you never see until the file is final.

So the formatting pass isn't one step — it's two. Clean the manuscript, then re-scan after conversion, because conversion is its own event with its own failure modes.

This matters because the fix for a submission portal is different from the fix for a KDP upload, which is different again from the fix for a print-ready PDF. If you're self-publishing, the KDP upload checklist after an AI-assisted first draft walks through the platform-specific landmines in more depth, and the free Kindle book formatter for Amazon KDP is built specifically to catch the conversion artifacts that slip through a manual pass — reflowed paragraph spacing, quote-mark corruption, and dash inconsistency that only appear once the file hits Kindle's rendering engine. It's worth running your manuscript through it even after you've done your own cleanup pass, simply because conversion bugs are a different species from drafting bugs. Pairing that with the KDP page count estimator also catches you off guard less often when your formatting changes shift your final page count right before upload.

If you're submitting to agents rather than self-publishing, the stakes are different but the principle holds: a formatting-consistent manuscript signals that you're someone who finishes things properly, which is exactly the signal an agent is looking for in the first five pages. Pair this final scan with a broader look at your opening hooks and first-page fixes guide, since agents will judge both mechanics and hook in the same first glance.

Where this fits in your actual process

This formatting pass isn't a replacement for developmental or line editing — it comes after. If you haven't already run a structural and line-level edit, do that first; the edit a book with AI guide and its companion piece on how to edit a book with AI cover that process in full. Formatting cleanup is the last mechanical pass before the manuscript leaves your hands, ideally after beta readers have weighed in — the beta reader workflow for AI-assisted manuscripts is worth running before this stage, since there's no point formatting a manuscript you're still substantially revising.

Once the manuscript is clean, mechanically and structurally, the rest of the submission or publishing process opens up — the publish, launch and distribution guide picks up exactly where this one leaves off. And if you're building your next project from scratch and want to avoid baking these inconsistencies in from the start, tools like the AI book outline generator and a solid write a book with AI workflow — paired with consistent use of a character consistency checker and fantasy worldbuilding tool if you're in that genre — tend to produce cleaner drafts in the first place, simply because the context stays unified instead of fragmenting across sessions and tools. For deeper technical reference on managing a single consistent manuscript inside one environment, the Entangled Text documentation covers the mechanics directly. And if cost is part of why you've been bouncing between tools and sessions in the first place, how to budget AI drafting for a full novel is worth a read before your next project, since a lot of formatting drift traces back to switching tools mid-draft to save money.

Set aside one focused hour before you submit anything. Run the audit prompt, build your style sheet, execute the mechanical pass, convert to your target format, and scan again. That hour costs you nothing creatively — it's not a rewrite, it's housekeeping — but it's the difference between a manuscript that reads as ready and one that reads as unfinished, regardless of how good the sentences inside it actually are.