Text Difference Analyzer
Compare two versions of your text. Changes are editing evidence, not watermark verification.
What the edit percentage measures
Word edit share is the total number of added and removed word occurrences divided by the total words in both versions. Replacing one word in two five-word versions gives 2 / 10 = 20%. A full replacement can reach 100%; punctuation-only edits can show 0% while still appearing in the comparison. This is a text-difference measure, not a detector confidence, removal success rate or proof of preserved meaning.
Words, sentences and formatting
The browser's English-locale Intl.Segmenter identifies word-like segments and sentence boundaries, including supported non-English writing systems. Results can vary across browser implementations. Word mode retains punctuation and whitespace changes. Sentence mode compares exact sentence spans, so a revised sentence can appear once as removed and once as added. Our segmentation is not the tokenizer used by an OpenAI or Google model.
An example that changes meaning
The example replaces a meeting time from nine to ten. The tool can show that edit, but it cannot judge whether the new time is correct or whether the revision preserves meaning. Review names, numbers, quotations and qualifications yourself. There is no percentage threshold that certifies TextGrain removal, human authorship or factual accuracy.
Privacy and comparison limits
Each version can contain up to 20,000 UTF-16 code units and 6,000 word, punctuation and space segments. A comparison stops with a specific error if it exceeds 2,000 token edits or the computation time limit. Inputs stay available so you can compare shorter passages. Processing is browser-local, with no text upload, persistence, sign-in or paid API. The engine uses jsdiff; it does not rewrite your text. Its license notice is retained. Optional consent-based usage events never contain the compared text or edit counts.