What prompt makes AI fix your grammar without rewriting you?
Paste this into Claude, ChatGPT, Gemini or Whizi and fill the brackets. It asks for the smallest fix and a record of each change.

Fix the grammar, spelling and punctuation in the text below. It is a [type of text, such as a cover letter or an essay paragraph] for [reader]. Change only what is wrong. Keep my word choices, sentence order and tone. Return the corrected text first, then a list with one line per change: the original, the fix and the rule. Don't touch names, numbers or quotes. Text: [paste your text]
Each sentence closes a gap. "Change only what is wrong" stops the rewrite. The list makes every edit something you can accept or refuse. The last rule guards the facts a model is most likely to alter without telling you.
Which grammar AI is free?
Grammarly, LanguageTool, Trinka, ChatGPT and Claude all have free plans, and each one stops you in a different place. The table lists what each vendor states on its own page as of November 2026. Where a page was silent, the cell says so.
| Tool | Free use | Sign up | Paid price | Best at |
|---|---|---|---|---|
| Grammarly | Free plan at $0: grammar, spelling and tone checks, 100 AI prompts a month | Not stated | Pro $12 a month billed yearly, $30 billed monthly | Checking inside Gmail, Docs and other apps |
| LanguageTool | 2,000 characters per text field, 3 AI paraphrases a day | Not stated | Not shown on the page when read | 30+ languages |
| Trinka | Basic lists 5,000 words of writing assistance and 5 AI writing requests a month | Not stated | Not shown on the page when read | Academic and technical writing |
| ChatGPT | Free plan lists "Limited messages with uploads" | Not stated | Go $8, Plus $20 a month (OpenAI launch post, read 22 November 2026) | Long text and following style rules |
| Claude | Free plan lists "Limited usage every 5 hours and every week" | Not stated | Pro $20 a month, or $17 a month billed annually | Careful edits with reasons |
| Whizi | 10 messages a day on Whizi AI, Auto and Claude Haiku, no card | Account needed | Pro $29.99 a month | A corrected Word file back, and two models on one text |
Whizi loses where the other tools are built for it. It doesn't sit inside Gmail or Google Docs and underline errors as you type, so Grammarly Free is the better pick for that. LanguageTool covers 30+ languages and needs no account for its browser add on. A chat app is the wrong shape for a one line email.
The checkers differ in what they promise. Grammarly's plans page puts full sentence rewrites and plagiarism checks on Pro. LanguageTool's premium page says Premium lifts the limit to 150,000 characters and adds more than 20,000 extra checks for seven languages. Trinka's pricing page lists 25 AI writing requests a month and editable file proofreading with tracked changes on Premium, against 5 requests and read only proofreading on Basic. The Google results for this query also include a Play Store app and a free checker on ryrob.com. I couldn't read either page for this table, so neither has a row.
What does a good grammar fix look like? Three worked examples
A good fix changes the errors and nothing else, then tells you what it changed. Each example below is written for this page, not the output of a named model, so treat the answers as the shape to look for.
Example 1, an email. Prompt: "Fix the grammar only in this email to my landlord. Keep my tone. List each change. Text: Me and my roommate has noticed the heater in the hallway are not working since monday, can you send someone to look at it." A good answer returns: "My roommate and I have noticed that the heater in the hallway has not worked since Monday. Can you send someone to look at it?" It lists the pronoun order, the verb that now matches "roommate and I", the capital on Monday, and the comma splice split into two sentences. It leaves "send someone to look at it" alone.
Example 2, an essay sentence. Prompt: "Correct only errors in this sentence from my history essay and explain each in a few words. Text: The treaty, which was signed in 1648 effected the borders of many european states." A good answer fixes "effected" to "affected", adds the comma after "1648", and capitalizes "European". It doesn't change "many" to "several" or add a claim about the treaty.
Example 3, a case where the model should ask. Prompt: "Fix the grammar in this line from my resume, and flag any edit that changes meaning. Text: Managed a team of 5 which increased sales." A good answer notices the ambiguity: did the team or the manager increase sales? It offers two fixes, one for each reading, and asks which is true instead of picking.
Can I get a corrected Word file back instead of pasted text?
Yes, in a chat app that makes Word files. The best order is to paste your text or attach the document, ask for the minimal edit prompt above, then ask for the result as a Word file with the change list on the last page. You open it, read the list, and decide.
This is where chat beats a checker for long documents. LanguageTool Free stops at 2,000 characters a field. A chat app takes a whole chapter. The trade is that it doesn't underline errors as you type, and it can alter meaning. Check the output with the read aloud step below.
What are the best prompts for grammar, spelling and punctuation?
Twelve prompts cover the jobs people bring to a grammar checker. Each one names a narrow job, since a narrow job returns a narrow fix.
Quick fixes
- Full pass: "Fix grammar, spelling and punctuation in the text below. Change only what is wrong and list each change with its rule. Text: [paste]."
- Spelling only: "Correct only spelling in the text below. Don't change grammar, punctuation or word choice. List each misspelled word and its fix. Text: [paste]."
- Punctuation only: "Fix only the commas, apostrophes and full stops in the text below. Show each changed sentence before and after. Text: [paste]."
- Tense check: "Find every tense shift in this text that isn't deliberate. List the sentence, the tense it uses and the tense the rest of the text uses. Don't rewrite anything yet. Text: [paste]."
Learning the rule
- Explain my mistakes: "Find the three grammar errors I make most often in this text. For each one, explain the rule in two sentences and give me two practice sentences to correct. Text: [paste]."
- Quiz me: "Write six sentences with a [rule, such as subject verb agreement or who versus whom] error in them for a [level] learner. Wait for my answer on each before giving the fix."
- Rule lookup: "When do I use a comma before 'which'? Explain with three correct and three incorrect examples, and name the style guide your rule comes from, such as Chicago or AP."
Work and school writing
- Formal email: "Fix the grammar in this email to [recipient]. Keep it under [number] words and don't add pleasantries I didn't write. Text: [paste]."
- Essay with a style guide: "Proofread this paragraph for [APA, MLA or Chicago] conventions on grammar and punctuation only. Don't comment on my argument. Text: [paste]."
- Non native writer: "I'm learning English and my first language is [language]. Correct this text and, for each error, say whether it is a common mistake for [language] speakers and why. Text: [paste]."
Edge cases
- Keep my voice: "Fix errors in this text but keep my short sentences, my contractions and my slang. Flag any fix that makes the text sound more formal. Text: [paste]."
- Second opinion: "Here is a text and a corrected version. Check the corrected version: did any edit change the meaning, a name, a number or a quote? List only the problems. Original: [paste]. Corrected: [paste]."
Prompt 12 is the safety net. Paste the original and the fix from any tool, Grammarly or a chatbot, and it checks the checker.
What mistakes ruin AI grammar fixes?
Four mistakes cause most bad results, and each has a quick fix.
- Letting the model polish. A bare "fix my grammar" invites a rewrite. The fix: say "change only what is wrong" and ask for a change list.
- Accepting every suggestion. A checker can be wrong about a name, a dialect or an intended fragment. The fix: read each change and reject what you don't agree with.
- Checking the corrected text but not the facts in it. Models can alter a figure or a quote while tidying. The fix: tell the model not to touch names, numbers or quotes, then compare them against the original.
- Pasting a long text into a short limit. A 2,000 character cap silently covers only part of an essay. The fix: split the text, or use a chat app for a long one.
A fifth one is quiet: a text you wrote in a different variety of English. Tell the model whether you want American or British spelling, or it will pick one for you.
Is AI good at grammar, and where does it go wrong?
AI is strong at spotting errors and weaker at leaving good sentences alone. In a 2023 evaluation, Fang and colleagues found that ChatGPT has "excellent error detection capabilities". The same abstract says 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."
In plain terms, a chatbot fixes your comma and also rewrites your sentence. That study tested an older model, so today's versions may differ, but the habit is why every prompt on this page says to change only what is wrong.
Models can also change facts while tidying prose. Why AI gets things wrong explains the mechanism. If your text is graded, read the school's rule on AI help first, and see does Grammarly count as AI for how one tool's features are treated. For feedback on an essay beyond grammar, AI essay grader covers what a rubric prompt can do.
One more limit: a grammar tool judges form, not truth. It will happily polish a wrong sentence.
Is there one app that fixes grammar and hands back a Word file?
Whizi is one app instead of a pile of single purpose tools, and it suits long text. On the website you can attach a PDF, Word or Excel file for the model to read, and scanned PDFs get a vision transcription pass. Word and Excel files made in chat have no plan gate, so a corrected draft can come back as a Word document on the free plan.
The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card. Pro carries Claude, GPT and Gemini in one app, and its first week is $1.99. Pro also has side by side Compare, which runs two models on one prompt, so you can run prompt 1 on two models and read only the edits they disagree on. Web search is on by default per chat, which helps when you ask a rule question such as prompt 7.
Grammarly still wins for checking as you type in other apps. If you write email and documents all day, keep it for that and use a chat app for the long drafts. For more on that tool, see is Grammarly free, and for the same job from another angle, AI proofreader and AI spell checker.
- Say who the text is for and ask for minimal edits
- Ask for a list of every change with its rule
- Tell the model not to touch names, numbers or quotes
- Read the corrected text aloud before you send it
- Split anything longer than the free limit of your checker
Frequently asked questions
Is there a free AI grammar checker?
Yes. Grammarly Free costs $0 with 100 AI prompts a month, LanguageTool Free checks 2,000 characters per text field and Trinka Basic lists 5,000 words of writing assistance, as of November 2026. ChatGPT and Claude have free plans with usage limits that the vendors don't state as numbers.
Is ChatGPT good at grammar checking?
It is good at finding errors and prone to over-correcting. A 2023 study found ChatGPT has excellent error detection but tends to rewrite beyond minimal edits. Ask it to change only what is wrong and to list each change, as in prompt 1.
Which grammar checker is best for students?
It depends on the length and the rules. Grammarly Free works inside other apps, LanguageTool covers 30+ languages, and Trinka aims at academic writing. Check your school's AI policy first, since some count generative rewrites as AI help.
Can AI fix grammar in a Word document?
Yes. Paste the text or attach the file, ask for minimal edits and a change list, then ask for the result as a Word file. Open it and read the changes yourself, because models can alter names, numbers and quotes.