Products SEO & GEO Agent Paid Ads Agent Website Agent Conversion Agent
How it works Pricing FAQ Blog Book a call

ChatGPT grammar correction: prompts that keep your voice

Updated · 6 min read

Short answer Tell ChatGPT to fix only errors: grammar, spelling, punctuation. Say it must not change your words, tone or sentence order, and must list every change with a reason. Without those limits it tends to rewrite whole sentences. Check the list, accept what is right, and keep client details out of the chat.

Want help applying this to your business? Talk it through on a free 30-minute call.

Book a call

What prompt fixes grammar without changing my writing?

Set limits before you paste anything. Copy this, then put your text between the two markers:

"Proofread the text between TEXT START and TEXT END. Fix only grammar, spelling and punctuation errors. Do not change my word choice, tone, sentence order or length. Do not add or remove ideas. Use US spelling. Return the corrected text, then a table with three columns: original, fix, reason. If a sentence is correct, leave it alone. TEXT START paste here TEXT END"

Each part does a job:

Part of the promptWhat it stops
"Fix only grammar, spelling and punctuation"Style edits you didn't ask for
"Do not change word choice, tone, order or length"Your voice turning into generic polished prose
"Use US spelling" (or Canadian)"Color" flipping to "colour", or the reverse
A table of original, fix, reasonHidden changes; you see every edit
"If a sentence is correct, leave it alone"Rewrites of sentences that had nothing wrong
Markers around your writingYour text being read as instructions

The markers follow OpenAI's own advice on marking where content starts and stops. Its prompt guide says XML tags "can help delineate where one piece of content (like a supporting document used for reference) begins and ends" (OpenAI API docs: Prompt engineering). Here the content is your draft, so a line like "please call me back" stays part of the email instead of becoming a request to ChatGPT.

Why does ChatGPT rewrite my style when I ask for a grammar check?

It is built to produce fluent text, and "fix this" sounds like permission to improve it. A 2023 study tested ChatGPT on standard grammar-correction test sets and found it "can freely correct errors to make the corrected sentences very fluent, possibly due to its over-correction tendencies and not adhering to the principle of minimal edits" (Fang et al., 2023).

Minimal edits is the editor's rule: change as little as needed to make the sentence correct. That study tested a 2023 model, so treat it as a warning, not a current score. The fix holds either way: name the rule in your prompt rather than hope for it.

OpenAI's help center makes the general point:

"Ensure your prompts are clear, specific, and provide enough context for the model to understand what you are asking." (OpenAI Help Center: Prompt engineering best practices for ChatGPT)

"Check my grammar" isn't specific. "Fix errors only, change nothing else" is.

How much should ChatGPT change? Pick a level

Decide before you ask. Most owner writing (quotes, emails, service pages) needs level 1.

LevelAsk forUse it for
1. ProofreadErrors only, list every changeClient emails, quotes, reviews replies, anything in your own voice
2. Light polishErrors plus sentences that are hard to follow; keep my words where possibleA page you wrote fast and want clearer
3. RewriteClearer, shorter, a set toneText you don't mind sounding different

For level 2, add: "Only reword a sentence if a customer could misread it. Mark those sentences separately." That keeps clarity fixes apart from error fixes, so you can say no to them.

How do I check what ChatGPT changed?

Read the change table, not the clean copy. A polished paragraph hides edits; a table shows them.

  1. Scan the reason column. "Subject-verb agreement" or "missing comma" is a real fix. "More engaging" or "smoother" means it broke the rule; reject it.
  2. Check tense and agreement across sentences. The 2023 study found ChatGPT could not reliably fix errors that span sentences, such as a tense shift from one sentence to the next.
  3. Check names, prices, dates and trade terms. These are the words ChatGPT is most likely to "correct" into something wrong, such as a product name or a local street name.
  4. Read the final text once out loud. If it no longer sounds like you, go back to your original and apply only the fixes you accepted.
  5. Push back in the same chat. "Undo change 3; that was my style" works. OpenAI calls this an iterative approach: "Start with an initial prompt, review the response, and refine the prompt based on the output" (OpenAI Help Center: Prompt engineering best practices for ChatGPT).

How do I stop retyping the prompt every time?

Put the rules in a project. Create a project named "Proofread", open its settings and paste the prompt rules as the project's instructions. OpenAI says:

"Project instructions apply only within that project and override your global custom instructions." (OpenAI Help Center: Projects in ChatGPT)

So your proofreading rules stay out of your other chats, and every new chat in that project starts with them. Add a short list of words never to change: your business name, service names, staff names, local place names.

What should I keep out of ChatGPT when proofreading?

Anything a client wouldn't want shared: health details, account numbers, full names next to private facts. Replace them with placeholders like [CLIENT] before you paste; the grammar check doesn't need them.

On personal plans, your chats may be used to train models unless you switch it off:

"On personal workspaces with ChatGPT Free, Plus, or Pro, data sharing is enabled by default." (OpenAI Help Center: Data controls in ChatGPT)

The switch is Settings, then Data controls, then "Improve the model for everyone". OpenAI says it does not, by default, use content from ChatGPT Business, Enterprise or Edu workspaces to train its models (OpenAI Help Center: Data controls in ChatGPT). Health practices have stricter rules; see Is ChatGPT HIPAA compliant?.

Example

A plumber's quote email

The draft says "We can be their tuesday, the part cost's $180 and labour is extra." The level 1 prompt returns three fixes: "their" to "there", "tuesday" to "Tuesday", "cost's" to "costs". It keeps "labour" because the owner changed the prompt to Canadian spelling. Without limits, ChatGPT might return "We'd be happy to visit on Tuesday," which is correct but no longer sounds like the plumber.

A salon's service page

The owner asks for level 2. ChatGPT fixes 6 errors and marks 2 long sentences as easy to misread, with shorter versions. She takes one and keeps the other.

A review reply

A one-line thank-you with one typo. Level 1 fixes the typo and nothing else. The reply still reads like the owner wrote it, because the owner did.

FAQ

Is ChatGPT good at grammar correction?

It finds errors well and explains them when asked. Its weak spot is changing more than needed, so your prompt must forbid style edits.

Can ChatGPT correct grammar in French or Spanish?

Yes, use the same prompt and name the language and region, such as Canadian French. How well it does varies by language; see ChatGPT supported languages.

Should I use ChatGPT or a grammar checker like Grammarly?

Grammarly checks as you type, in "more than 1 million websites and apps" (Grammarly: How Grammarly works). ChatGPT suits longer pasted pieces you want explained.

Does the free version of ChatGPT do grammar checks?

Yes. The prompt above needs no paid features. Projects depend on your plan and workspace settings (OpenAI Help Center: Projects in ChatGPT).

Different for you?

Keep learning

NextChatGPT grammar checker: a step-by-step way to proof draftsA six-step routine to proofread a full draft with ChatGPT: prepare it, fix errors by section, review each change, check consistency, then read it yourself.

Ask us one question

One line is enough. We reply by email.

Sources

  1. Fang et al. (2023): Is ChatGPT a Highly Fluent Grammatical Error Correction System?
  2. OpenAI Help Center: Prompt engineering best practices for ChatGPT
  3. OpenAI API docs: Prompt engineering
  4. OpenAI Help Center: Projects in ChatGPT
  5. Grammarly: How Grammarly works
  6. OpenAI Help Center: Data controls in ChatGPT