ChatGPT Prompts for Editing a Novel and Getting Feedback (Claude and Gemini Too): Keep Your Voice and Get More Than Praise
When you ask ChatGPT, Claude or Gemini to look at a novel you wrote, the trick is to narrow it down to one thing you want checked and say up front, "Don't rewrite anything, just point out problems." This article gives you ChatGPT prompts for editing a novel and getting feedback in four flavors (reader impressions, a structural critique, line editing and proofreading), shows how to keep your own voice, and explains how each company handles your data so you can check before pasting an unpublished manuscript.

Key takeaways
- Ask for one thing at a time: reader impressions, structure, line editing or proofreading
- Start with "Don't rewrite, just point out problems." Do the actual fixing yourself
- Don't say "I wrote this," and ask for the same number of strengths and concerns, with quotes
- Before pasting an unpublished manuscript, check the setting that decides whether it's used for training
The basics of asking AI to edit your novel or give feedback
When you ask ChatGPT, Claude or Gemini to look at a novel you wrote, the trick is to narrow it down to one thing you want checked and say up front, "Don't rewrite anything, just point out problems." If all you say is "edit this," you may get the whole manuscript polished from top to bottom, or a list of nothing but compliments, and either way it's hard to tell what to fix next.
Think of a visit to the doctor: "my right knee hurts on the stairs" gets you much further than "check me for anything wrong." A manuscript works the same way. It's easier to choose how to ask if you think in terms of these four kinds of feedback.
| What to ask for | What it looks at | When it fits |
|---|---|---|
| Reader impressions | How a reader feels (where they got pulled in and where they stopped) | Right after you finish a draft, when you want a sense of how the whole thing lands |
| Critique (structure) | Pacing, why characters act the way they do, foreshadowing | Once the plot is settled, when you want to check the story's skeleton |
| Line editing (prose) | Hard-to-read sentences, repetition, word choice | Once the skeleton is solid, when you want to polish the sentences |
| Proofreading | Typos and missing words, inconsistent spellings, contradictions with earlier chapters | Right before you show it to anyone or post it |
The recommended order is top to bottom, because fixing the structure can wipe out sentences you polished earlier.
6 steps to get AI to edit your novel
The steps come in three stages: prepare, ask and revise.
- 1Decide on one thing you want checked
- 2Add brief information about the story
- 3Describe the style to keep and the sentences not to touch
- 4Ask for feedback only, first
- 5Choose what to fix and get suggestions one spot at a time
- 6Revise it yourself and have the AI read it again
1. Decide on one thing you want checked
Pick one from the table above: reader impressions, critique, line editing or proofreading. If you ask for "plot, prose and typos, all at once," the answers come back shallow and spread thin. Ask for the second thing separately, after you've fixed the first.
2. Add brief information about the story
In a few lines, write the genre, the intended readers, which chapter this is and the feeling you want readers left with (for example: bittersweet, but hopeful). Say what you're aiming for up front, so that, for example, a horror story's unsettling aftertaste doesn't get "fixed" as a flaw.
3. Describe the style to keep and the sentences not to touch
Attach a "style note" covering point of view, your habitual phrasing and sentence endings, sentence length and how much dialogue you use. If you mark sentences you love with "don't change this one," they'll be left out of the suggestions.
4. Ask for feedback only, first
At first, ask only for "where and why something bothers you." If you get a rewritten manuscript first, you'll wear yourself out just hunting for what changed.
5. Choose what to fix and get suggestions one spot at a time
Pick only the feedback you agree with, and ask something like "give me three ways to fix this one spot." Use the suggestions as material for deciding which direction to take the fix.
6. Revise it yourself and have the AI read it again
Using the suggestions as a reference, rewrite it in your own words. Then have the AI read the revised version and check only whether the earlier problems have been resolved.
ChatGPT prompts for editing a novel and getting feedback (copy and use)
Replace the ___ parts to fit your story, then paste your manuscript at the end. They work in ChatGPT, Claude and Gemini.
Prompt for reader impressions (more than just praise)
Read the following novel manuscript as an ordinary reader who regularly reads fiction, and tell me your impressions. [ABOUT THE STORY] Genre: ___ / Intended readers: ___ / Chapter ___ of ___ [WHAT TO WRITE] 1. How you feel right after finishing, in one sentence 2. Two places that pulled you in (quote the sentence and explain why) 3. Two places where you stopped reading or got confused (quote the sentence and explain why) 4. Whether you'd want to read the next chapter (yes/no, and why) [RULES] List the same number of strengths and concerns. Don't rewrite the text. [MANUSCRIPT] (paste here)
Prompt for a critique of structure and plot
You are a fiction editor. Look only at the "structure and plot" of the following manuscript. You don't need to comment on the wording. [ABOUT THE STORY] Genre: ___ / The feeling I want readers left with: ___ [OVERALL SYNOPSIS] (in 3 lines) [WHAT TO LOOK AT] - Places where the pacing is too fast or too slow - Places where a character's reason for acting is missing - Foreshadowing that never pays off, and places that over-explain [HOW TO ANSWER] Up to 3 concerns, most important first. For each, one line each for "where," "problem" and "direction for the fix." Don't give me rewritten sentences. [MANUSCRIPT] (paste here)
Prompt for line editing (keep your own voice)
Use this prompt after you've received feedback in step 4 and decided what to fix (step 5).
In the following manuscript, suggest revisions only for the sentences listed below. Don't change my writing style. [MY STYLE] Point of view: ___ / Mostly past tense / I like stringing short sentences together / Not much dialogue [SENTENCES TO FIX] (one spot at a time, chosen from the feedback) 1. "___": what bothers me (e.g. the meaning is hard to follow) 2. "___": what bothers me [RULES] - Don't change anything except the sentences listed - In your suggestions, keep the original words and phrasing as much as possible [HOW TO ANSWER] A table with the columns "original / suggestion / reason" [MANUSCRIPT] (paste the scene here)
Proofreading prompt for typos, inconsistent spellings and contradictions
Proofread the following manuscript. You don't need to judge whether the writing is good or bad. [WHAT TO LOOK AT] - Typos and missing words - Inconsistent spellings (e.g. "Ken" vs. "Kenny," "gray" vs. "grey") - Contradictions with earlier chapters (names, what characters call each other, dates, belongings, injuries) [ESTABLISHED FACTS FROM EARLIER CHAPTERS] (names, what characters call each other and setting details, as bullet points) [HOW TO ANSWER] A table of what you found with the columns "where (which paragraph) / current text / issue." Don't give me corrected text. [MANUSCRIPT] (paste here)
Tips for asking: avoid over-editing and keep your voice

Get more than praise
An AI's feedback can be pulled toward the feelings of the person asking. A 2023 paper from Anthropic's research team (Towards Understanding Sycophancy in Language Models) reports that when users add "I really like this text" or "I wrote this," the AI's evaluation tends to lean more positive, and when they add "I don't like it," it tends to lean more critical.
There are three ways to counter this.
- Don't say "I wrote this" or "this is my best work": just write "the following manuscript" and hand it over.
- Ask for the same number of strengths and concerns: setting a number makes it more likely you'll get concerns too.
- Have it answer with quotes: asking "quote the sentence you mean" makes it harder to get feedback that could apply to any manuscript.
Avoid over-editing
Spell out the scope of changes, as in "don't rewrite" or "only change the sentences I listed." If you get the answer as a table with "original / suggestion / reason," you can throw out just the suggestions you don't like.
Keep your own voice
An AI's suggestions may produce tidy sentences, but your personal quirks won't necessarily survive. Attach your style note every time and add "keep the original words and phrasing as much as possible." Read the revised sentences out loud to make sure they don't stick out.
Don't give instructions that contradict each other
When we went back through the story requests submitted to Original Novel (22 requests on November 6–7, 2026), 7 of them contained instructions that contradicted each other. For example, asking for "a story complete in one chapter" while also wanting a twist ending and several foreshadowing payoffs. With so few requests this is only a rough guide, but the same kind of contradiction can happen when you ask for edits.
| Problem | Likely cause | Line to add to your prompt |
|---|---|---|
| Nothing but praise, so you can't tell what to fix | You added "I wrote this" / you didn't set a number | "The same number of strengths and concerns, with quotes" |
| Your personal quirks disappear | You didn't describe your style | "My style is ___. Keep the original words and phrasing" |
| Too much feedback to know where to start | You didn't narrow down what to look at | "Up to 3, most important first" |
| Mixed answers to "make it shorter" and "add more description" | Your instructions contradict each other | "If in doubt, prioritize ___" |
Another option is to show the same manuscript to two AIs and start with the spots both of them flagged.
The limits of AI editing: what not to take at face value
- Matters of taste: "This metaphor is hard to follow" is one opinion. Decide by weighing it against what you were aiming for.
- The whole of a long manuscript: in a long exchange, the AI can miss things from the early part. For a long novel, hand it over one chapter at a time, along with the established facts from earlier chapters.
- Fact-checking: for feedback about historical accuracy, medicine, law, place names and so on, don't decide based on the AI's answer alone. Check it against sources.
- Rules for posting and contests: if you plan to submit an AI-edited manuscript to a story site or a contest, check its AI rules first. We've summarized the rules of major story-posting sites in Japan in How to write a novel with AI.
What happens to the manuscript you paste? What the three AI companies officially say
Before you paste an unpublished manuscript, check whether the text you enter is used to train the AI. The table below reflects what we confirmed in each company's official help on November 11, 2026 (it may differ by plan, region or age).
| AI | Setting that controls training use | Chats not used for training | Things to watch out for |
|---|---|---|---|
| ChatGPT | "Improve the model for everyone" under "Data Controls" in Settings. On by default for personal use of Free, Plus and Pro. Turning it off doesn't delete your chat history | Temporary Chat: not used to improve the models and not saved to your history. May be kept for up to 30 days for safety purposes | If you send a rating with 👍 or 👎, the entire conversation may be used for training even if the setting is off |
| Claude | The model-improvement switch in the Privacy section of Settings (Privacy → "Help improve our AI models"). Conversations are used for improvement in cases such as when you allow it | Incognito chat: not used for model improvement, even if the model-improvement setting is on | If you allow it, new conversations may be kept in a form separated from your identity for up to 5 years for training. Conversations flagged as problematic in a safety review may be used for safety measures |
| Gemini | A separate item inside "History" (chat content, including text and uploads). Off by default. When it's on, your chats are used to train generative AI (History must be on to turn it on) | Temporary Chat: not used to train AI. Kept in your account for about 2 months | Some conversations are read by human reviewers. Google's official help advises against entering confidential information you wouldn't want a reviewer to see. Even with the setting off, if you send feedback, the feedback and related conversations from the past 24 hours are collected |
When you want feedback on text you'd rather not share, such as a manuscript you're about to submit to a new-writers' prize, the safe choice is to turn the training setting off or use a chat that isn't used for training, and not press the rating buttons. Work and school accounts may follow different rules, and setting names can change, so check each company's official help too.
Read AI-written stories with a reader's eye
If you read other people's stories while paying attention to "where I got pulled in and where I stopped," you'll also sharpen your eye for your own manuscript. Original Novel (original-novel.com) is a fiction site launched in November 2026 where you can create stories with AI and read them.
Read AI-written stories with a reader's eye
See more: Read AI-written stories with a reader's eye ›Stories written by AI on this site. As you read, pay attention to where you got pulled in and where you stopped.
The Only Voice I Can't HearFor teens · Showbiz & idolsCompletedMay We Begin With a Lie?For adults · RomanceSummary
When you ask ChatGPT, Claude or Gemini to edit your novel or give feedback, narrow it down to one thing, get the feedback first, and do the fixing yourself. If you don't say "I wrote this" and you ask for the same number of strengths and concerns, each with a quote, you're less likely to get nothing but praise. Before you paste an unpublished manuscript, also check the setting that decides whether it's used for training.
Sources (date checked)
- Data controls in ChatGPTOpenAI Help Center / Checked Nov 18, 2026
- Is my data used for model training?Anthropic Privacy Center / Checked Nov 18, 2026
- How long do you store my data?Anthropic Privacy Center / Checked Nov 18, 2026
- How do I change my model improvement privacy settings?Anthropic Privacy Center / Checked Nov 18, 2026
- Gemini Apps Privacy HubGoogle Gemini Help / Japanese / Checked Nov 18, 2026
- Manage and delete your Gemini Apps activity (computer)Gemini Apps Help (Google) / Japanese / Checked Nov 18, 2026
- Towards understanding sycophancy in language modelsAnthropic / Checked Nov 18, 2026
FAQ
What's a good ChatGPT prompt for editing a novel?
Pick one thing you want checked (impressions, structure, line editing or proofreading), add a little information about the story and your writing style, and start by asking, "Don't rewrite anything, just point out what bothers you." Once you've decided what to fix from that feedback, list only those sentences and ask for suggestions in a table with the columns "original / suggestion / reason." For sentences you love, write "don't change this one."
Why does AI feedback on my writing come back as nothing but praise?
An AI's assessment can be pulled toward the feelings of the person asking. A 2023 paper from Anthropic reports that when users add "I really like this text" or "I wrote this," the AI's evaluation tends to lean more positive. Don't say you wrote it, and ask for the same number of strengths and concerns, each with a quote. That makes the bias easier to keep in check.
When AI edits my writing, my voice disappears. What can I do?
Write a style note covering point of view, your habitual phrasing and sentence endings, sentence length and how much dialogue you use, attach it every time, and add "keep the original words and phrasing as much as possible." Limit the scope too: "only change the sentences I listed." Don't use the suggestions as they are; rewriting them in your own words is the safest way.
Is the novel manuscript I paste into ChatGPT used to train the AI?
It depends on your settings. OpenAI's official help says that for personal use of Free, Plus and Pro, "Improve the model for everyone" is on by default and can be turned off, and that Temporary Chats are not used to improve the models (as of November 11, 2026). If you send a rating with 👍 or 👎, the entire conversation may be used for training even when the setting is off.
Can I use the same prompts to get feedback from Claude or Gemini?
Yes. The prompts in this article work as they are in ChatGPT, Claude and Gemini. Hand over a long manuscript one chapter at a time, and check how your data is handled in Claude's privacy settings or in Gemini's "History" setting.



