How to Use AI to Edit Your Novel: Prompts and Workflow for Self-Editing a First Draft (2026 Guide)
12 min read · Updated October 2026
You have a finished first draft, and now comes the part that decides whether the book is any good. AI can help with that job, but only if you give it the right tasks. It is useful for spotting patterns, checking consistency and giving you a fast, tireless first reader. It is poor at judging taste, protecting your voice or knowing what your book is trying to be. This guide gives you a pass-by-pass workflow, prompts you can copy, the limits you should respect, and a clear view of when a human editor is still worth paying for. It fills the gap between a revision checklist and a tool roundup: what to actually type, in what order, and what to do with the answers.
What can AI actually do when editing a novel?
Think of AI as an assistant editor with a good memory for what you paste in and no stake in your book. It is strong at reading a scene and telling you what it notices: repeated words, pacing that sags, a character who says the same thing three times, a timeline that does not add up. It is also good at producing lists, summaries and comparisons quickly, which is most of what self-editing needs.
It is weaker at the things that make a novel work for readers. It cannot feel suspense or tension the way a reader does, it tends to flatter, and it often pushes prose toward a smooth, generic register. Left alone, it will happily sand down the odd phrasing that makes your voice yours. Use it to diagnose, and keep the decisions for yourself.
It also has a limited working window. Very long manuscripts may not fit in one conversation, and even when they do, attention to detail in the middle can slip. Work chapter by chapter or scene by scene, and keep your own notes as the source of truth. Tools and limits change often, so check what your chosen model currently handles.
How should you prepare your draft before using AI?
Do not start with AI. Leave the draft alone for a while if you can, then read it yourself once, ideally on a different device or as a printout, and note what bothers you. Your own reaction is the most useful data you have, and it stops the AI from setting the agenda. If you have not yet done a structural pass, the revision checklist in our guide to revising a novel is the right first step.
Then build a short reference sheet to paste in with every request: genre, target reader, point of view and tense, a two-paragraph summary, a list of main characters with one line each, and your voice rules, such as short sentences, British spelling, no adverbs on dialogue tags. This keeps feedback anchored to your book rather than to an average novel.
Finally, think about privacy. Check the data and training settings of whatever service you use before pasting in unpublished work, and keep a backup of every version. Save your original draft untouched and work on copies.
What is a good workflow for AI-assisted self-editing?
Edit from big to small, and never polish sentences in a scene you may cut. A workable order is structure, then characters and continuity, then scene-level craft, then line edits, then proofreading. Each pass has a different question, and AI helps most when you ask only one question at a time.
- Structure: give it your chapter-by-chapter summary and ask where the pacing stalls, which scenes repeat a beat, and where the stakes are unclear.
- Characters and continuity: check motivation, names, ages, timeline and setting details against your reference sheet.
- Scene craft: dialogue, sensory detail, point of view slips, showing versus telling.
- Line edit: repetition, filler words, rhythm, clichés.
- Proofread: typos, punctuation and consistency, then a final human read.
After each pass, make the changes yourself in your own manuscript. Paste the revised text back only if you want a second check. This keeps the words yours and gives you a clear record of what changed.
Which prompts work for structure and continuity?
Good prompts state the role, give the context, name one task and say what format you want back. For structure, try: “You are a developmental editor for [genre]. Here is my reference sheet and a one-line summary of each chapter. Identify the three weakest points in the pacing, scenes that repeat the same dramatic beat, and any place where the protagonist’s goal becomes unclear. Do not rewrite anything. Give your reasoning for each point.”
For continuity, paste a chapter and your reference sheet and ask: “List every factual detail in this chapter about characters, places, dates and objects, then flag anything that contradicts the reference sheet or earlier details I have pasted. Return a table with the detail, the location in the text and the conflict.” Building the table first, then checking it, is more reliable than asking whether there are any errors.
For character arcs, ask for evidence rather than opinion: “Quote the lines in these scenes that show what [character] wants, what they fear and how they change. Tell me where the change feels unearned.” If you want to go deeper on tracking this kind of detail, a story bible makes every one of these prompts stronger.
Which prompts work for line editing without losing your voice?
The safest line-editing prompts ask for diagnosis, not rewriting. For example: “List every word or phrase I repeat more than three times in this chapter, every filler word such as just, really and suddenly, and every sentence that begins the same way as the one before. Quote each one. Do not change the text.” You then fix them in your own style.
When you do want suggestions, constrain them: “Suggest a tighter version of these five sentences only, keep my vocabulary and rhythm, and give two options each.” Ask it to preserve dialect, slang and deliberate fragments. Compare its version against yours, and take a change only when you can say why it is better.
Watch for the usual traps. AI tends to add adverbs and stock phrases, flatten distinctive dialogue into the same polite voice for every character, and invent details to fill gaps. If a suggested line sounds like it could appear in any book, reject it. A reading level checker can also tell you whether your prose has drifted from what your genre readers expect.
What are the limits and risks of AI editing?
The main risks are false confidence and homogenised prose. AI can say a chapter is strong when readers would find it dull, and it can miss a plot hole that depends on something said two hundred pages earlier. It may also state a correction confidently that is simply wrong, such as a grammar rule or a factual detail, so verify anything you are not sure of in a style guide or dictionary.
Be careful with the line between editing and generating. Using AI to analyze and suggest is different from letting it write large passages that end up in your book. Retailers such as Amazon have their own rules about AI-generated content and disclosure, and those rules change, so read the current policy; our guide to KDP and AI disclosure explains the distinction.
Do not treat AI feedback as a substitute for readers. It cannot tell you whether a joke lands, a twist surprises or an ending satisfies. Beta readers who love your genre will catch what a model cannot, and a long-term author’s career is built on those readers.
When should you still hire a human editor?
Hire a human when the book is going to be sold and you want it to compete. A developmental editor brings judgment about genre expectations, emotional payoff and what to cut, and a copyeditor or proofreader gives a trained final check. AI self-editing makes the manuscript cleaner and your own thinking sharper, so the professional’s time is spent on deeper problems rather than repeated words.
Costs vary widely with length, genre and the editor’s experience, so get quotes rather than trusting a single figure. An editing cost estimator can help you plan a budget, and the CruxOne marketplace lets you hire editors, designers and other professionals if you want somewhere to start. CruxOne is an AI book publishing platform for indie authors, used in a web browser, with AI writing and editing tools alongside it.
A sensible sequence is: your own read, AI-assisted passes, revisions, beta readers, then a professional edit and proofread. If budget is tight, prioritize by need. A structural problem is worth fixing with a developmental editor before you pay anyone to correct commas.
Frequently asked questions
Can AI edit a whole novel in one go?
Not reliably. Long manuscripts can exceed what a model handles well, and detail gets missed. Work chapter by chapter with a reference sheet.
Will AI editing make my writing sound generic?
It can if you accept its rewrites. Ask for diagnosis and flagged problems, then fix them in your own words and keep your voice rules in every prompt.
Is it safe to paste my unpublished novel into an AI tool?
Check the service’s data and training settings first. Use options that do not train on your content where available, and keep your own backups.
Does AI replace a human editor?
No. It is a useful first reader for patterns and consistency, but it cannot replace an editor’s judgment about story, genre and reader experience.
Do I have to disclose AI editing when I publish on KDP?
Rules differ between AI-generated content and AI-assisted editing and they change over time. Read Amazon’s current policy before you publish.