Grammar Checker

What the checker is instructed to do

The request carries one fixed instruction: fix every grammar, spelling, punctuation and basic clarity mistake while preserving the original meaning, tone, formatting and language, without changing the writing style, adding or removing content, or explaining anything. The model answers with the corrected text only. Randomness is set very low (temperature 0.2), so the same input gives almost the same correction every time.

The correction is generated by a language model hosted on Groq's platform, reached through this site's server. The text is therefore processed by a third-party service rather than on your device, and the site itself keeps nothing. The model is told to answer in the language it was given, which is how Arabic, Spanish or French text comes back in the same language.

Kinds of mistakes it corrects

What it won't do

• It won't list the errors or explain them. You get the clean text, not a report, so you learn by comparing the two versions.
• It won't flag weak style, long sentences or passive voice; the instruction forbids rewriting. For that, use the Paraphrasing Tool or the AI Humanizer.
• It won't check facts, names or figures. A wrong date in correct grammar stays wrong.
• It may “fix” deliberate choices: dialect, song lyrics, a character's broken English in fiction, or an error you quoted on purpose.
• It may switch between British and American spelling in a text that mixes them, since the instruction asks for consistency of language, not of variety.
• Markdown, code and placeholders like {name} are supposed to be preserved, but check them; a model can tidy a bracket it took for punctuation.

Limits on length and frequency

The input is trimmed to 5,000 characters on the server, and the counter under the box turns red when your text is longer than that. In English that's about 800–900 words. The corrected reply is capped at 1,500 tokens, which covers 5,000 English characters with room to spare, but languages that use more tokens per word, such as Arabic or Japanese, can reach the cap on a maximum-length paste and lose the last lines.

A long essay is best checked a section at a time, which also makes the comparison easier. The site accepts 12 AI requests per minute from one address; beyond that it answers “Too many requests” until the minute rolls over.

Reviewing the correction efficiently

Because there's no highlighting, the fastest review is a diff. Paste the original into the left side of the Diff Checker and the corrected text into the right; every changed word is marked, and you can accept the fixes you agree with and ignore the ones that touched something intentional.

If the model rewrote a sentence more than you wanted, the “basic clarity” part of the instruction is why. Paste that sentence back on its own and it will usually return a lighter correction, because there's less context inviting a reword.