ChatGPT Grammar Checker: How to Proofread Well With ChatGPT

Yes, ChatGPT works as a grammar checker: paste your text, tell it to fix spelling, grammar and punctuation, and ask it to list every change so you can review them. The catch is that it is a general-purpose assistant, not a dedicated proofreader, so you get the best results by limiting how much it rewrites and checking its work.

This guide covers the prompts that work, how editing has changed since OpenAI replaced canvas with writing blocks, what research says about ChatGPT’s accuracy, and the privacy settings worth checking first. Details are current as of November 2026; ChatGPT’s interface changes often, so menus and feature names may shift.

How to check grammar with ChatGPT (step by step)

1. Open a new chat at chatgpt.com or in the ChatGPT app. A fresh chat keeps earlier instructions from leaking into the edit.
2. Paste your instruction first, then your text. Put the text below a clear separator such as three dashes or quotation marks so ChatGPT knows where your instructions stop.
3. Say how much change you want. “Fix only grammar, spelling and punctuation. Do not rephrase for style” produces very different output from “Improve this paragraph.”
4. Ask for a list of changes. Without one, it can be hard to see what ChatGPT altered in a long passage.
5. Review each change and accept or reject it yourself. Ask follow-up questions (“Why did you change ‘which’ to ‘that’?”) whenever a fix looks doubtful.
6. Paste the final version back into your document, and run your word processor’s own spell check as a last pass.

Prompts that work

For minimal, proofreader-style corrections:

Proofread the text below. Correct only objective errors in spelling, grammar and punctuation. Do not change word choice, tone or sentence structure unless a sentence is grammatically wrong. After the corrected text, list every change as “original → corrected” with a one-line reason.

For learning, when you want to understand your mistakes:

Find the grammar mistakes in the text below. For each one, quote the sentence, explain the rule in plain language, and show the corrected version. Do not rewrite anything else.

For a specific style or dialect:

Proofread this for a formal business audience using British English spelling and the Oxford comma. Flag anything that is ambiguous rather than guessing what I meant.

Naming the audience, tone and spelling convention (US or UK English, for example) removes a lot of guesswork and reduces unwanted changes.

Writing blocks replaced canvas

If you’ve read older guides, they probably tell you to open canvas and use the Suggest edits shortcut. When OpenAI introduced canvas in October 2024, Suggest edits added inline comments to a draft, a lot like a reviewer’s notes.

That has changed. According to OpenAI’s ChatGPT release notes, writing and coding moved directly into chat responses through writing blocks and code blocks on May 28, 2026, and canvas is no longer available in GPT-5.5 Instant or GPT-5.5 Thinking. A June 8, 2026 update let longer writing open in a focused full-screen editor. Zapier’s canvas overview also notes that writing and code blocks replaced canvas. Third-party reports say paid users could still reach canvas on some older legacy models for a limited time. We could not confirm from OpenAI how long that will last, so don’t build a workflow around it.

Using writing blocks for proofreading

OpenAI’s help article on writing blocks says ChatGPT may place a draft it expects you to revise or reuse in a writing block. Inside a block you can:

• edit the text directly yourself
• ask ChatGPT to revise selected text or the whole draft
• undo or redo recent AI-assisted edits
• open the draft in a full-screen view and copy it out

Edits save with the conversation after a short delay, so follow-up requests work on your latest version. In a Temporary Chat, OpenAI says edits may not persist after the conversation ends. Available actions vary by plan, device, workspace settings, model and rollout, so you may not see every option.

The practical workflow: ask ChatGPT to return your corrected text in a writing block, then select a single sentence or paragraph and ask for a targeted fix. That gives you finer control than asking it to rewrite everything at once. Some users on OpenAI’s community forum have complained that writing blocks lack the inline suggestion comments canvas had. For now, the “list every change” prompt above is the most reliable way to see exactly what was altered.

How accurate is ChatGPT at grammar?

The most cited independent evaluation is the 2023 arXiv paper Is ChatGPT a Highly Fluent Grammatical Error Correction System? by Tao Fang and colleagues. Its findings still match what many writers notice:

• It catches a lot. In human evaluation, ChatGPT left fewer errors uncorrected than the other systems compared, including Grammarly.
• It over-corrects. The authors found a clear tendency to over-correct and not to follow the principle of minimal edits. Its output reads fluently, but partly because it rewrites more than necessary.
• Cross-sentence errors are harder. The paper reports weaker performance on document-level problems such as agreement, coreference and tense consistency across sentences.

Keep in mind that this study tested an early 2023 version of ChatGPT, and today’s GPT-5.5 models are different. We haven’t found an equally rigorous public benchmark for the current models, and the many “ChatGPT vs Grammarly” scorecards online mostly don’t publish their methods, so treat their percentages with caution. The over-correction pattern is the main thing to watch for, and the minimal-edit prompt above is designed to curb it.

ChatGPT vs a dedicated grammar checker

Many writers use both: a dedicated checker or built-in spell check for routine typing, and ChatGPT when they want an explanation, a second opinion on a tricky sentence, or a change of tone.

Privacy: check this before pasting sensitive text

Proofreading often means pasting work emails, contracts or unpublished drafts. According to OpenAI’s data controls FAQ:

• On personal Free, Plus and Pro plans, conversations can be used to train OpenAI’s models by default.
• You can opt out in Settings → Data controls → “Improve the model for everyone.” On iOS and Android, open the sidebar, tap your profile icon, then go to Settings and Data controls. Opting out does not delete your chat history.
• Giving thumbs-up or thumbs-down feedback on a response can still make that whole conversation eligible for training.
• Temporary Chats are not used to improve models, though OpenAI says they may be kept for up to 30 days for safety purposes.
• Content from ChatGPT Business, Enterprise and Edu workspaces is not used for training by default.

Even with training turned off, follow your employer’s or school’s rules on AI tools, and remove names, account numbers and other personal details before you paste.

Tips for better results

• Work in chunks. For long documents, proofread a section at a time, and review each list of changes before moving on.
• Lock your terms. Tell ChatGPT which brand names, technical terms or quoted passages it must leave untouched.
• Ask for a consistency pass. Because cross-sentence errors are a known weak spot, run a separate prompt such as “Check this for consistent tense, number agreement and pronoun references across paragraphs.”
• Don’t accept blindly. ChatGPT can sound confident about “errors” that are really style preferences. If a rule surprises you, check a style guide.
• Watch academic rules. If you’re a student, check whether your institution allows AI proofreading and whether you need to disclose it.

Used this way, ChatGPT is a capable proofreader that explains its fixes. Use it as a second reader that you double-check, not something you trust to fix everything on its own.