Claude's watermark, gone. Your meaning, intact.

Paste the text. See the clean version, every number and name checked.

Up to 100,000 words per order, files up to 15 MB. Free: the first part, about 600 words, processed and shown; the rest once you pay.

No account, no subscription.

How the Claude watermark works, in two minutes

The mark goes in while Claude chooses its words, so there is nothing in the file to find and delete: it lives in the wording itself.

It is a statistical watermark of the SynthID family, and we rebuilt that technology on our own bench, with our own key. That is how we take the mark out, and how we test every remover that promises to: the same marked text, the same detector, every run published.

Every run is published.

Who brings us work

You are probably here with one of these

The job is the same in all four: the mark comes out, the meaning stays, and the report shows you which is which. We do not ask what the material is or where it came from.

After the run

What you get back: four checkable things

The measurement

One marked text, 32 facts of 32 kept

How the mark works, on our own bench

A sentence we wrote and marked ourselves, with our own key. The tinted words are the ones the mark steered the wording towards.

The results indicate that the method delivers a substantial improvement in overall accuracy on every test.

our detector, holding our own key: mark found

On 5 September 2026 one marked sample of 1,029 words went through our processing and through the best known open remover. Both came out in the same band on the mark; ours kept all 32 figures, names and terms of 32, the open rewrite 28. Measured on our own bench, with our own key, and published with the sample.

Test any remover yourself. Take a text carrying a mark of ours, run it through whatever you want to try, and our detector says what came back. Open the tester

Price

One payment for this document. No subscription.

$5

$5 for a document up to 10,000 words. Longer documents are priced by length; you see the exact price before paying.

Everyone around here sells a monthly plan. We do not: you are far more likely to pay a stranger once for one file than to sign up for a month you will forget to cancel.

Start with the free preview

Questions people actually ask

Frequently asked

Did you really take the watermark off?

The mark lives in the wording, and the wording is re-expressed from the first line to the last, with every number and name checked against your original. On our own bench, where we hold the key, the detector finds nothing after the run. We also keep your result, so the day Anthropic opens its detector we run the archive through it and show you the numbers.

Every run, published with the sample.

Can the Claude watermark be bypassed rather than removed?

People search for this as bypass or workaround, and it is the same work. The mark is placed while the text is generated, in the words the model chose, so there is nothing in the file to switch off: the wording itself is re-expressed, from the first line to the last, and the report shows what survived of your meaning. Anyone promising you a certificate is promising something Anthropic keeps to itself; what you can do is take a marked sample of ours and put any remover to it, this one included.

Test any remover on our marked sample.

Why not just strip the invisible characters?

The mark is not made of characters. Anthropic places it as the text is generated, so clearing characters out leaves it where it was: we ran a free character cleaner over marked texts ourselves, and the mark was found in every one of them afterwards.

What each free path returned.

Is this legal?

The law requires the model provider to mark what its model writes; it asks nothing of the person using it. We process material that belongs to you.