Free · No sign-up · Runs in your browser
Modern language models hide a statistical watermark in the words they choose. This tool finds the few segments that carry almost all of that signal, and rewrites them with natural synonyms — so your text keeps its meaning and loses its fingerprint.
Paste your text or upload a file. The analyzer scores every segment and isolates the ones with the heaviest watermark weight.
This runs once per session and keeps automated traffic from draining the detection budget.
The highlighted segments carry the signal. Rewrite them and watch the score drop.
The detector’s probability that this document carries a watermark. 50% is pure chance. This is the same number the before/after bar reports, so re-analysing a cleaned text gives back exactly its “after” score.
These segments carry a disproportionate share of the watermark signal. Breaking them collapses the detector's statistic for the whole document.
Highlighted spans are the load-bearing segments. Everything else is statistically neutral.
Each targeted segment is rewritten several ways, every version is re-scored, and only the version that lowers the score is kept. Stubborn segments get another round with more words in play.
Watermark-carrying words replaced with meaning-preserving synonyms. The figures below are the detector’s score for the whole document, before and after.
How to read these numbers: this is a client-side heuristic detector. It reproduces the standard green-list z-test used by watermarking research, but without the generator's private key, so scores are a statistical estimate of watermark pressure — not a verdict from any specific vendor's detector. Use the tool on text you own or are permitted to edit.
The method
Watermarks are not spread evenly. A handful of token choices carry most of the statistical evidence — so that is where the tool works.
Your text is split into segments and each token is hashed against its predecessor to determine green-list membership — the same construction used in watermarking literature. Each segment gets a z-score.
The three to five highest-scoring segments are highlighted. Together they typically account for the majority of the document's detectable signal while covering a small fraction of its length.
Inside those segments only, several rewrites are generated using part-of-speech-matched synonyms. Every version is scored again, and only the one that actually lowers the score survives. Stubborn segments get another round with more words in play.
Why this tool
Most rewriters shred your whole document. This one edits a few dozen words and leaves your writing intact.
Only 3–5 segments are touched. Structure, tone, terminology and argument stay exactly as you wrote them.
Green-token rate, z-score and signal concentration are computed live — not decorative progress bars.
Part-of-speech matching, capitalization retention, and a stop-word blacklist prevent nonsense substitutions.
No accounts, no logs, no uploads. Your document is processed in the browser tab you are reading this in.
Every substitution is listed with its original word, so you can review or reverse anything you dislike.
Drag in a .txt or .md file, and copy or download the cleaned result in one click.
Answers
A statistical bias deliberately embedded while a language model generates text. At each step the vocabulary is pseudo-randomly split into a preferred (“green”) list and a non-preferred list based on a hash of the previous tokens, and the model is nudged toward the preferred list. The output reads naturally, but a detector holding the same key can measure the skew and flag the text.
It rebuilds that green-list construction as a proxy detector, scores every segment of your text with a z-test, ranks them, selects the 3–5 heaviest, and rewrites content words inside those segments using synonyms from the free Datamuse lexical API. Because the green list depends on the previous token, a single substitution re-rolls the list for the words that follow it.
Yes. No account, no trial, no usage cap. The site is static and the synonym dictionary it queries is a free public API, so there is nothing to charge for.
No. Analysis and rewriting happen locally. The only outbound requests are single-word lookups to the public synonym dictionary — never sentences, paragraphs or files. Turn off the “Online synonym API” switch and the tool becomes fully offline, using its built-in dictionary.
It is deliberately conservative: content words only, part of speech matched, names, numbers, acronyms and function words skipped, and a hard cap of a few substitutions per segment. Every change is logged for review. Always re-read the result before you use it.
Version 1 targets English. The detection maths is language-agnostic, but the synonym source is English only, so rewriting quality outside English is not guaranteed.
Use it on text you own or are authorised to edit, and follow the rules of whatever institution or client you are writing for. This tool is a text-processing utility; it is not a licence to misrepresent authorship.
No sign-up, no upload, no cost. Paste it in and see the signal drop.
Open the tool