Proofread text without rewriting it
Spelling, grammar and punctuation errors only, each quoted with its correction — awkward-but-correct sentences are left exactly alone.
Last updated
Fill in the blanks
Assembled in your browser — nothing you type is stored or sent anywhere.
Your prompt
Model-agnostic — works in Claude, ChatGPT, Gemini or any AI chat.
Why this prompt works
- →Scoping the task to mechanical errors only, and explicitly forbidding restyling, stops a proofreading request from turning into an uninvited rewrite.
- →Requiring the exact original quote alongside the correction makes every fix checkable against the source, instead of trusting a wholesale "corrected" version.
- →Permission to report zero errors matters because a proofreader under pressure to find something invents phantom issues in text that was already clean.
Part by part
- Find only mechanical errors
- The scope, stated first. Proofreading and editing are different jobs, and a model asked to proofread will edit unless told not to.
- Do NOT rewrite, rephrase, restyle or improve the writing in any way
- Four verbs rather than one, because each is a separate way the boundary gets crossed. The emphasis is doing real work here.
- if a sentence is awkward but not incorrect, leave it alone
- Resolves the specific judgement call. Awkward-but-correct is where a proofread turns into an edit, and the instruction removes the ambiguity.
- quote the exact original text
- Makes each correction locatable and checkable. A list of corrections without their context cannot be applied confidently.
- do not invent one just to seem thorough
- Models are reluctant to return nothing. Explicit permission to find no errors is what makes a clean result trustworthy.
Common mistakes
Proofread this.Returns a rewritten version with the errors fixed and the style changed. You then have to diff it to find out which changes were corrections and which were opinions.
Check this for errors and improve it.Combines two jobs, so the improvements bury the corrections. If you want both, ask separately — proofread first, then decide whether to edit.
Fix any mistakes in this text.Mistake is broad enough to include stylistic choices, so a deliberate sentence fragment or an intentional repetition gets corrected out. Mechanical is the word that draws the line.
Questions
Why insist so hard on not rewriting?
Because it is the default behaviour and it is hard to undo. Once a model has rewritten a paragraph, separating the two corrections from the fifteen stylistic changes takes longer than proofreading it yourself.
Will it catch everything?
No. It is reliable on spelling, agreement and punctuation, and it misses errors that require knowing what you meant — a correctly spelled wrong word, or a consistent misuse of a name. A human pass is still worth it on anything published.
Why does it invent errors?
Because returning nothing reads as unhelpful, so it finds something to say. Explicit permission to report a clean result is the only reliable fix, and it is why that line is in the prompt.
Does it handle style guides?
Only if you name one. Serial commas, quotation placement and date formats differ between guides, so without a named standard the corrections follow whichever convention it defaults to.