Does Anthropic really watermark Claude's text?

Yes, since 11 August 2026. Every Claude model launched on or after 2 August 2026 embeds an imperceptible mark in generated text, applied at the model level so it is present on every Claude surface, and applied worldwide rather than only in the EU. The driver is EU AI Act Article 50, which became enforceable on 2 August 2026. Anthropic states the mark travels through copy-paste and may persist through some editing.

Can the watermark actually be removed?

Anthropic's own documentation says the mark may not survive heavy editing, paraphrasing, translation, mixing with other content, or very short passages. Paraphrasing is on that list, and the rewrite here is a paraphrase pass. But Anthropic has not published its detection mechanism, so nobody outside Anthropic - including us - can currently verify whether a mark is present or gone. We will not claim a result we cannot demonstrate. When a detector is published we will test against it and publish what we find.

What does a detected mark actually prove?

Less than people assume. Anthropic states a mark indicates content may have been processed by Claude - not that Claude authored it. Text you wrote yourself and asked Claude to proofread carries the same mark as text Claude drafted from scratch. The mark cannot distinguish those two cases.

What are the two layers you clean?

Character-level and statistical. The character layer is zero-width codepoints, Unicode tag characters, variation selectors, exotic spacing and typographic punctuation - removed by exact codepoint matching, verifiable by re-scanning, no model involved. The statistical layer is the vocabulary and cadence patterns, plus whatever the watermark contributes; that can only be addressed by rewriting.

Can invisible characters really be hidden in text?

Yes, and they survive ordinary copy-paste. Zero-width characters, Unicode tag characters and variation selectors all render as nothing while remaining fully present in the data. Run the free scan on any text and you will see the exact count.

Does the rewrite change my meaning?

It is instructed not to. Facts, figures, names, dates, quotations, citations, URLs, paragraph structure and claim strength are all held fixed, and any sentence without a marker is returned verbatim. Read the result before you use it — that is true of any tool that touches your writing.

Will this defeat a specific AI detector?

We make no claim about any named detector, and no claim to verified watermark removal. Detectors are proprietary, they change without notice, and they disagree with each other. Anthropic's watermark has no public detector at all yet. What we state precisely: the character-level markers are removed exactly and verifiably, and the wording markers in your scan report are what the rewrite targets.

Why is the scan free?

Because it costs us almost nothing. It is exact character matching, not a model call, so there is no reason to charge for it. You only spend words when you ask for an actual rewrite.

What happens to my text?

It is processed to produce your result and saved to your history so you can retrieve it later. You can delete any entry. It is never used to train a model.

How is usage counted?

By word, against the input text. Scans are free and never counted. Each plan also caps the words in a single request, so long documents need a higher tier or splitting up.

Can I use this on text I wrote myself?

Yes, and plenty of people do. Human writing picks up invisible characters from PDFs, CMS editors and word processors, and the typographic punctuation cleanup is useful on its own. The scan will tell you whether there is anything there.

How to remove ChatGPT's watermark from text

OpenAI researched a text watermark and did not ship it. What that means for GPT output, what is genuinely in it, and what a 'ChatGPT watermark remover' is actually removing.

Removing the watermark · 3 min read

Start with the part that changes the whole question: as of this writing, OpenAI has not deployed a text watermark in ChatGPT.

OpenAI built one. Reporting and the company's own statements describe a working system held back rather than shipped, for reasons that were partly commercial and partly technical — the same paraphrasing weakness that limits every sampling-based watermark, plus the risk of disadvantaging non-native English writers who use the tool to edit.

So if you are searching for a ChatGPT watermark remover, the honest framing is that there is probably no statistical watermark in your text to remove. There is still plenty in it worth cleaning.

What is actually in GPT output

Hidden characters. Not a deliberate mark, but they are there — copied in from training data, emitted by the tokenizer, or introduced when text passes through a web interface. Zero-width characters, non-breaking spaces, occasionally variation selectors.

Typographic punctuation, heavily. Em dashes as a mid-sentence connector, curly quotes and apostrophes instead of ASCII, the single-codepoint ellipsis U+2026. This is the most reliable signature in GPT text and it is entirely countable.

Vocabulary and cadence. delve, leverage, robust, seamless, tapestry, pivotal. Sentence-opening Moreover and Furthermore. Empty hedges like "it is important to note that". Sentence after sentence of near-identical length.

None of that is a watermark. All of it is identifiable, and the last two categories are what an AI detector is actually responding to when it flags GPT text.

So what do "ChatGPT watermark removers" remove?

The character layer, in almost every case. That is a genuinely useful thing to do — it just is not watermark removal, because there is no watermark.

This is worth knowing before you pay for one. A tool promising to strip OpenAI's watermark from your text is promising to remove something that has not been deployed. The cleanup it performs is real; the framing is not.

What is worth doing

Strip the characters. Exact, verifiable, and it fixes practical problems beyond provenance: hidden characters break search, diffs, code blocks and spreadsheet imports. Re-scan afterwards and the count should be zero.

Fix the punctuation. Em dash to a spaced hyphen, curly quotes to straight, U+2026 to three periods. This is the single strongest typographic tell and normalising it is safe — no meaning changes.

Rewrite the vocabulary and rhythm, if your concern is a detector rather than a watermark. Detectors measure perplexity and burstiness: how predictable the word choices are and how much sentence length varies. That is a style judgement, not a provenance reading, and it is what actually drives a flag on GPT text.

Be clear about which problem you have. They call for different work.

If OpenAI ships one later

The EU AI Act's Article 50 became enforceable on 2 August 2026 and requires providers to mark machine-generated output. Anthropic responded by marking Claude's text globally. Google had already shipped SynthID.

Whether OpenAI follows is not something we can tell you today, and the answer may already have changed by the time you read this. If it does ship, expect the same properties every sampling-based watermark has: distributed across the passage, invisible in the codepoints, degraded by paraphrase and translation, undetectable in short text, and readable only by the key holder.

The character-stripping advice above would be unaffected. It never addressed a statistical watermark in the first place.

What we do

The character strip is exact — every invisible codepoint and typographic substitution, matched directly and removed. Re-scan and confirm zero.

The rewrite pass clears vocabulary and cadence markers while holding facts, structure, register and length steady. Against GPT text, that is aimed at detector heuristics and at how the writing reads, not at a watermark, because there is not currently one to aim at.

We would rather say that than sell you removal of a thing that does not exist.

Run a free scan to see exactly what is in your GPT text.

How to remove ChatGPT's watermark from text

What is actually in GPT output

So what do "ChatGPT watermark removers" remove?

What is worth doing

If OpenAI ships one later

What we do

Related

How to remove Gemini's watermark from text

Can you remove SynthID from text?

Does Claude's API output carry the watermark?