Watermark Removal Test Report

A monthly record of what was tested, on which models, at what length, and where removal still failed. The methodology is published so the numbers can be reproduced or argued with.

Short answer

A monthly record of what was tested, on which models, at what length, and where removal still failed. The methodology is published so the numbers can be reproduced or argued with.

What is actually true about watermark removal

Why publish this

Vendor claims about removal are unverifiable. Publishing sample sizes and failures makes the claim falsifiable, and the report is the only page in this niche a reader can audit.

Known failure modes

Very short text, legal and technical boilerplate, code, and heavily constrained formats all flatten under any rewrite. They are reported as failures, not excluded.

How to reproduce

Paste the same source text, choose the same mode, and compare the before and after scores. The detector code path is deterministic for a given input.

Does the watermark survive?

Measured behaviour for this kind of mark. Blanks would be guesses; these are not.

OperationResultWhy
Copy and paste into another appSurvivesThe token choices travel with the characters.
Light editing or swapping a few synonymsSurvivesOnly tokens you actually replace stop carrying the bias.
Full paraphrase or re-generationRemovedEvery marked token is replaced.
Machine translation and backMostly removedMeaning survives; the statistical pattern largely collapses.
Screenshot or print to PDFSurvivesA text watermark is never visual.
Uploading to a CMSSurvivesThe platform stores your characters as-is.

How to remove it

  1. Paste the text. Paste the watermark removal passage into the field at the top of this page. One anonymous removal is included.
  2. Run the check. Check the score first so you have a baseline. The check is free and needs no account.
  3. Choose paraphrase. Paraphrase preserves quality. Back-translate is the heavier option when you need a bigger change.
  4. Run the removal. Every token is re-generated, so the sampling bias the watermark rides on is replaced.
  5. Compare the scores. A large drop is the result you want. A flat score means the passage was too short or too formulaic to carry much signal.
  6. Proof and copy. Check names and numbers, then copy the clean text out.

Why use Zuro

Zuro vs. the alternatives

MethodSpeedQualityCostAccount
ZuroUnder 10 sNo quality lossFree check, 20 for $5, $29/yrAnonymous first run
Asking watermark removal to rewrite itSecondsOften re-marked by the same keyFreeRequired
Manual rewriting10-30 minDepends on skillFreeNo
Synonym swapperSecondsLeaves most tokens intactFreeUsually
Detector onlySecondsReports, does not removeVariesUsually

Frequently asked questions

Is removing the watermark removal watermark free?

The score check is free and needs no account, and the first removal is anonymous. After that a free account includes 3 removals, a 20-removal pack is $5 one-time, and unlimited use is $3.99/month or $29/year.

Does removing the watermark removal watermark reduce quality?

No. Every token is re-generated so meaning, tone and length stay intact; only the word choices the mark lives in change.

Do I need an account?

Not to check a score, and not for the first removal. An account is only needed for more than one removal, so you can judge the output before signing up.

Are my files stored?

Text is processed in memory. Logged-in removal sessions are stored so they appear in your dashboard, and you can log out at any time.

Will light editing remove the watermark?

No. Anthropic says light editing probably will not, and a complete rewrite where every word is replaced will. That is the method Zuro automates.

How is this different from an AI detector?

Detectors guess from style and have no watermark key. Zuro shows that same kind of score, breaks it down, and then actually removes the signal.

What formats are covered for AI-generated text?

Text, pasted documents and PDF body text are handled inline. For images and video, the page covers metadata stripping and inpainting.

Accuracy and limitations

Related

Watermark Removal Test Report — condensed

A monthly record of what was tested, on which models, at what length, and where removal still failed. The methodology is published so the numbers can be reproduced or argued with.

Watermark survival:

Zuro: free score check, one anonymous removal, 3 free with an account, 20 for $5, unlimited $29/year. API $0.01/1k words. Canonical: https://zuro.cc/watermark-report

Tools for agents: https://zuro.cc/mcp · https://zuro.cc/llms.txt