Content Writing · Editing & voice
The Best Prompt for Editing and Proofreading
The best prompt for editing tells the model how far it may go and what it must leave alone. This one defines three levels (proofread, copyedit, line edit), has the model describe the author's voice before touching the text, and requires it to query anything that affects meaning. You get the edited draft, a table of changes with reasons, and a list of questions for the author.
Editing and Proofreading Prompt
Edits a draft at the level you choose, protects the author's voice, and returns a change log plus queries instead of silent rewrites.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Intermediate
- Complexity
- Multi-part reply
- Variables
- 5
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
5 variables · fields marked * matter most
# Role You are a copy editor working on someone else's writing. Your job is to make the draft cleaner and clearer while it still sounds like the person who wrote it. The author should read the result and recognize their own piece. # The job - Level of edit: [Level of edit] - Who will read the piece: [Who will read it] - Style rules to follow: [Style rules] - Leave these alone: [Leave these alone] Levels, so we mean the same thing: - Proofread: fix typos, spelling, punctuation, grammar errors and inconsistencies. Change nothing else. - Copyedit: proofread, plus fix unclear sentences, wrong word choices, repetition and inconsistent terms. - Line edit: copyedit, plus tighten wordy passages and improve flow between sentences and paragraphs. Structure and argument stay as they are. # Draft <draft> [Draft to edit] </draft> # How to work 1. Read the whole draft before changing anything. Write three or four observations about the author's voice: sentence length and rhythm, level of formality, characteristic words or constructions, how they handle humor or emphasis. These are what you will protect. 2. Edit at the level requested and no further. If you see a problem above that level, such as a paragraph in the wrong place during a proofread, raise it as a query instead of fixing it. 3. Before each change that is not a clear error, ask whether it fixes a problem for the reader or only makes the sentence sound more like you. If the second, leave it. Fragments, unusual word choices, one-sentence paragraphs and sentences that begin with a conjunction are choices unless they cause confusion. 4. Keep the author's vocabulary. Do not swap a plain word for a more formal one, or a vivid one for a safe one. 5. Never change meaning, facts, names, numbers or quoted material. If something looks wrong or ambiguous, query it. # Constraints - Do not add new ideas, examples, or a concluding sentence the author did not write. - Follow the style rules above where given. Otherwise follow whichever convention the draft uses most often and make it consistent. - Keep the length close to the original unless a line edit calls for cuts. # Output format 1. Voice notes. 2. The edited draft, complete, with no commentary inside it. 3. Change log: a table with the original text, the revised text and the reason. Group identical fixes into one row. 4. Queries for the author: anything you chose not to change but think they should look at.
Use it when you are
- Proofreading an article, report or newsletter before it is published
- Copyediting a colleague's or client's draft without rewriting it in your own style
- Tightening a wordy piece while keeping what makes it sound like its author
- Cleaning up writing in a second language while preserving the writer's phrasing where it is clear
Why this prompt works
The level of edit is defined, not implied
Asked to edit, a model tends to rewrite. Spelling out what proofread, copyedit and line edit each include gives it a boundary and gives you a shared vocabulary. The rule that problems above the chosen level become queries means you still hear about a misplaced paragraph without having it moved for you.
Voice is described before anything is changed
Telling a model to preserve voice is vague until it has said what the voice is. The voice notes make it commit to specifics, such as short declaratives or dry asides, which it then has to edit around. You can also correct the notes if it has misread the author.
A test separates fixes from preferences
The question in step 3, whether a change helps the reader or only makes the sentence sound like the editor, is the judgment a good human editor applies. Listing fragments and conjunction openers as deliberate choices removes the usual targets of over-correction, which is how distinctive prose gets flattened into safe prose.
Every change has to be accounted for
The change log makes the edit reviewable. You can accept or reject by row instead of comparing two versions by eye, and a reason column quickly shows when a change was a matter of taste. Queries keep questions of meaning with the author, where they belong.
How to use it
- Pick the lightest level that will do the job. If you are unsure, start with a copyedit.
- Add house style rules if you have them, and list anything the editor must not touch.
- Paste the full draft so the model can judge voice and consistency across the whole piece.
- Read the voice notes first. If they miss something important about how you write, say so and ask for the edit to be redone.
- Review the change log row by row and tell the model which changes to revert.
- Answer the queries yourself; do not ask the model to resolve questions about your facts or meaning.
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | For long drafts, the reply may be cut off before the change log. Ask it to continue from where it stopped, or edit the piece in sections of a few hundred words. |
| Claude | Expected to work | The draft sits inside XML-style tags, which Anthropic's guidance recommends for keeping long pasted text separate from instructions. Do not remove them. |
| Gemini | Expected to work | If the edit goes further than the level you chose, quote the level definition back and ask it to revert changes outside that scope. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: A bookshop owner who writes a chatty weekly newsletter wants it cleaned up without losing the tone. Illustrative inputs only.
# Role You are a copy editor working on someone else's writing. Your job is to make the draft cleaner and clearer while it still sounds like the person who wrote it. The author should read the result and recognize their own piece. # The job - Level of edit: Copyedit - Who will read the piece: Regular customers of an independent bookshop - Style rules to follow: UK spelling - Leave these alone: The opening 'So.' Levels, so we mean the same thing: - Proofread: fix typos, spelling, punctuation, grammar errors and inconsistencies. Change nothing else. - Copyedit: proofread, plus fix unclear sentences, wrong word choices, repetition and inconsistent terms. - Line edit: copyedit, plus tighten wordy passages and improve flow between sentences and paragraphs. Structure and argument stay as they are. # Draft <draft> So. The delivery finally came, three days late and missing a box, which if you've ever ordered anything in November you'll know is basically on time. Inside: the new paperbacks, a calender nobody ordered, and and forty copies of a cookbook I'm fairly sure we asked for four of. Its fine. We like cookbooks. Come in Saturday, their will be biscuits, and I will try to sell you one. </draft> # How to work 1. Read the whole draft before changing anything. Write three or four observations about the author's voice: sentence length and rhythm, level of formality, characteristic words or constructions, how they handle humor or emphasis. These are what you will protect. 2. Edit at the level requested and no further. If you see a problem above that level, such as a paragraph in the wrong place during a proofread, raise it as a query instead of fixing it. 3. Before each change that is not a clear error, ask whether it fixes a problem for the reader or only makes the sentence sound more like you. If the second, leave it. Fragments, unusual word choices, one-sentence paragraphs and sentences that begin with a conjunction are choices unless they cause confusion. 4. Keep the author's vocabulary. Do not swap a plain word for a more formal one, or a vivid one for a safe one. 5. Never change meaning, facts, names, numbers or quoted material. If something looks wrong or ambiguous, query it. # Constraints - Do not add new ideas, examples, or a concluding sentence the author did not write. - Follow the style rules above where given. Otherwise follow whichever convention the draft uses most often and make it consistent. - Keep the length close to the original unless a line edit calls for cuts. # Output format 1. Voice notes. 2. The edited draft, complete, with no commentary inside it. 3. Change log: a table with the original text, the revised text and the reason. Group identical fixes into one row. 4. Queries for the author: anything you chose not to change but think they should look at.
What a good response should contain
- Voice notes that pick up the short dry sentences, the direct address and the understatement
- Clear errors fixed: calender, the doubled 'and', Its, their
- The opening fragment, 'basically on time' and the conversational rhythm left as written
- A change log listing each fix with a short reason
- A possible query about whether 'one' in the last sentence means a biscuit or a cookbook, left for the author since the ambiguity may be the joke
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- Models can miss errors and occasionally introduce them. Proofread the final version yourself before publishing, especially names and numbers.
- It cannot fact-check. A wrong date or figure will only be queried if it looks inconsistent within the draft.
- The change log may not capture every small change. If accuracy matters, compare the versions with a diff tool.
- This prompt deliberately avoids structural editing. If the piece needs reorganizing, outline it again before editing sentences.
Alternative prompts
Content Writing
Brand Voice Guide Prompt
Analyzes your existing writing samples and derives a usable voice guide: traits with evidence, tone by situation, word lists and before-and-after rewrites.
- ChatGPT
- Claude
- Gemini
Content Writing
Blog Post Writing Prompt
Turns your own notes and experience into a focused blog post with one clear argument, and marks every spot where it needs a real example or a fact check.
- ChatGPT
- Claude
- Gemini
SEO
Content Refresh Prompt
Audits an existing page section by section, decides whether to refresh, rewrite, merge or leave it, and lists what to keep, update, cut, add and verify.
- ChatGPT
- Claude
- Gemini
- Perplexity
Related resources
Guide
How to Improve AI Responses
A diagnostic approach to weak AI answers: name the fault, trace it to the gap in your prompt, then follow up precisely or rewrite and start again.
Technical guide
How to Test AI Prompts
A seven-criterion rubric and a nine-field test record for evaluating prompts, plus why this site describes results in words and publishes no scores.
Frequently asked questions
Why does AI editing so often make writing sound generic?
With no instruction about scope, a model moves text toward the most common phrasing it knows, which removes the unusual choices that made the writing yours. Limiting the level of edit, naming the voice features to protect and requiring a reason for each change are the three controls this prompt uses to counter that.
Which level should I choose?
Proofread when the piece is final and you only want errors caught. Copyedit when sentences are sometimes unclear or terms are inconsistent. Line edit when the draft is sound but wordy. If you are editing someone else's work, agree on the level with them first.
Can I use it on a very long document?
Yes, in sections. Run the first section, keep the voice notes, and paste them into the style rules field for later sections so the edit stays consistent. Editing a whole long document in one reply increases the chance of a truncated answer or an incomplete change log.
Sources and references
- Prompt engineering overview — Anthropic
- Prompt engineering guide — OpenAI
Was this prompt useful?
No account needed. Do not include personal information.