You've written a message to your landlord, a post for the neighborhood group, a report for your boss, a note for your kid's teacher. It says what you mean, but you're not sure about a comma, a verb tense, or that word you always misspell. And reading it a fifth time, you no longer see anything.

An AI assistant can catch a lot of those mistakes in seconds. The trap is that if you just say "correct this", it often hands back a different text: smoother, more formal, full of words you'd never use. This guide shows you how to proofread your writing with ChatGPT, Claude, Gemini or any other assistant, step by step, so you fix the errors and keep your own voice. If you've never used an assistant, start with our guide on how to use ChatGPT as a beginner.

What the AI won't do for you

Let's be clear from the start:

  • It doesn't know what you meant. If a sentence is ambiguous, it picks one meaning and "fixes" it in that direction, which may not be yours.
  • It doesn't check your facts. A wrong date, a wrong amount, a misspelled name of a person or a company: the grammar can be perfect and the content still false. It can even introduce a new error while rephrasing, with a confident tone (that's what we call a hallucination).
  • It isn't always right about grammar. It's often very good, but it can flag something that was correct, or miss a real mistake, especially on rare rules, regional usage or specialized vocabulary.
  • It doesn't decide for you. Every change is a suggestion. The final text, and your name at the bottom, are yours.

Think of the assistant as a quick, patient second pair of eyes: very useful, but not the last word.

Before you start: what not to paste

Proofreading means pasting a whole text, and texts often contain more than you notice. Before you paste, take thirty seconds to remove or replace:

  • Contact details: full names, home addresses, phone numbers, email addresses, account or policy numbers.
  • Other people's private information: a colleague's health, a neighbor's conflict, a child's school situation. Replace names with "my colleague" or "[NAME]".
  • Confidential work content: client names, internal figures, unreleased projects, anything covered by a confidentiality agreement. If your employer has rules about AI tools, follow them.

You can put the real names back after the check: they don't change the grammar. If you're not sure what counts as sensitive, play a round of Safe to paste?, a free game where you spot the personal details to remove before pasting a text into an AI chat. For the bigger picture, read ChatGPT and your personal data.

Step 1: decide what kind of check you want

"Correct my text" can mean three very different things. Choose before you ask:

  • Proofreading: spelling, grammar, punctuation, typos. Your words and sentences stay the same. This is what you want most of the time.
  • Light editing: the same, plus clumsy sentences, repetitions and unclear passages, flagged as suggestions.
  • Rewriting: a new version, with a different tone or length. Useful sometimes, but then it's no longer really your text.

A simple rule of thumb: the more personal the text (a thank-you note, a message to a friend, a letter that has to sound like you), the more you should stick to pure proofreading. Light editing makes more sense for practical texts where clarity matters more than style, like instructions for a babysitter or a notice for your building. Keep rewriting for the rare cases where you really want a different text, and then reread it as if someone else had written it.

Also note who the text is for and what it's for: a quick message to a friend doesn't need the same care as a letter to an administration.

Step 2: paste your text with clear limits

The secret is to tell the assistant what it must not do. Here's a prompt to copy and adapt; the bracketed parts are yours to fill in:

Proofread the text below. Fix only spelling, grammar, punctuation and typos.
Do not change my wording, my tone or the structure of my sentences.
Do not add or remove information. Keep [American / British] English.
Context: it's [an email to my landlord about a repair].

Text:
"""
[PASTE YOUR TEXT HERE]
"""

The three quotation marks mark clearly where your text starts and ends, so the assistant doesn't confuse it with your instructions. The spelling variant matters: without it, the assistant may "correct" words that were right where you live.

If you write in a second language, say so: "English isn't my first language; keep my simple style, just make it correct." That one sentence avoids a lot of over-polished rewrites.

Step 3: ask for a list of changes, not just a new version

A corrected text alone hides what changed. You might miss a modification you don't agree with. Ask for the changes in a table instead:

Don't rewrite the whole text. Give me a table with one row per change:
the original passage, the correction, and a short reason
(spelling, grammar, punctuation...).
If you're not sure about a correction, say so in the reason column.

Now you see every change and why. You'll often notice patterns: the same comma mistake, the same confusion between two words. That's how you actually improve, instead of depending on the tool.

If your text is long, work section by section: paste a few paragraphs at a time. It's easier to review, and the assistant is less likely to skip anything.

Step 4: accept or refuse each change yourself

Go through the table row by row. For each change, ask yourself: is it really a mistake, or just a different way of saying it?

  • Clear errors (a typo, a missing letter, a verb that doesn't agree): accept.
  • Style changes disguised as corrections (a word swapped for a "better" one, a sentence turned around): refuse if your version was correct and sounds like you.
  • Doubtful cases: ask a follow-up, or check a dictionary or a grammar reference you trust.
For change number [4], is my original version actually wrong,
or is your version just a different style? Explain simply.

Then apply the changes you accepted in your own document, by hand. It takes a few more minutes than copying the assistant's version, but nothing slips in without your consent.

Step 5: a second pass for clarity, as suggestions only

Once the errors are fixed, you can ask for help on clarity, still without handing over the pen:

Now read the corrected text as [my landlord] would.
List up to 5 places where a sentence is unclear, too long or repetitive.
For each, suggest one alternative, but keep my vocabulary as much as possible.
Don't rewrite the rest.

Pick only the suggestions that help. If you use the same rules every time (your spelling variant, "don't change my tone", "always give me a table"), save them in your custom instructions so you don't retype them.

Step 6: do the final read-through yourself

The last check is yours, and it's the one that matters most:

  • Read the text out loud. Your ear catches missing words and heavy sentences that your eyes skip.
  • Check names, dates, numbers and amounts against your source: the email you received, the contract, the calendar. No proofreading tool knows whether "March 12" is the right date.
  • Check the facts if the text makes claims. For a method, see how to fact-check an AI answer.
  • Put the real names back if you replaced them before pasting.

For important documents (a cover letter, a resume, an official letter), consider also asking a person you trust to read it. Our guides to a cover letter with ChatGPT and writing an email with AI go further for those cases.

Common pitfalls

  • Saying only "correct this": you'll often get a rewrite, not a proofread. State what to fix and what to leave alone.
  • Copying the new version without comparing: changes you'd refuse slip in unnoticed. Ask for a table of changes.
  • Forgetting the spelling variant: "color" and "colour" are both right, depending on where you write.
  • Pasting sensitive details: names, addresses and account numbers don't help the grammar. Remove them first.
  • Believing the AI on every rule: it can be wrong about grammar too. When in doubt, check a dictionary or a reference you trust.
  • Skipping the final read: the tool doesn't know your facts, your dates or your intention. Your read-through is the real last step.

Mini exercise: proofread a real text in 15 minutes

  1. Choose (2 min): pick a real text you need to send this week, a few paragraphs long. Remove names, addresses and numbers, or replace them with [NAME].
  2. Proofread (3 min): use the prompt from step 2, with your spelling variant and the context.
  3. Table (3 min): ask for the table of changes from step 3.
  4. Decide (4 min): go through every row. Accept, refuse, or ask a follow-up on doubtful ones. Apply your choices by hand in your own document.
  5. Read out loud (3 min): read the final text aloud, check every date, number and name against your source, and put the real names back.

What to notice: how many changes were real errors and how many were just style. And which mistake comes back most often in your writing: that's the one worth learning.

Going further

To understand why an assistant writes such fluent, correct-looking sentences, and still sometimes slips in a confident mistake, play the free chapter "Light the lantern" in your browser, no account needed: play the chapter. To practice giving precise instructions like "fix this, don't touch that", try the Lab, and to sharpen your eye for errors stated with confidence, play a round of Hallucination Hunt.

If you're writing in another language, our guide to translating with AI shows how to get a translation that's right, not just readable. And if a term trips you up, the glossary is there.

Frequently asked questions

Can ChatGPT check grammar and spelling?

Yes, it's one of its most useful everyday uses: it catches many spelling, grammar and punctuation mistakes quickly. It isn't infallible, though. It can miss an error or flag something that was correct, so treat each change as a suggestion and check doubtful cases in a dictionary or a grammar reference.

How do I stop ChatGPT from rewriting my whole text?

Say explicitly what it may and may not change: "fix only spelling, grammar and punctuation; don't change my wording, my tone or my sentence structure." Asking for a table of changes instead of a new version also keeps you in control of every modification.

Is it safe to paste my text into an AI to proofread it?

Remove what the grammar doesn't need first: full names, addresses, phone numbers, account numbers, other people's private information and confidential work content. You can put them back after the check. If your employer has rules about AI tools, follow them for work documents.

Is using AI to proofread considered cheating?

Checking spelling and grammar in your own text is generally treated like using a spell checker. Rules differ for schoolwork, exams and some professional settings, so check what's allowed where you are. Keeping your own words and deciding every change yourself is also what keeps the text honestly yours.