AI text watermark remover for Claude, GPT and Gemini

Yes, and Anthropic says so openly. Its support documentation states that Claude embeds an imperceptible watermark directly into the text it generates, and attaches signed C2PA provenance metadata to image files. This covers Claude models launched on or after 2 August 2026 across the API, Claude, Claude Code, Cowork and Claude Tag, including AWS, Google Cloud and Microsoft Foundry. Anthropic frames it as a transparency commitment under Article 50(2) of the EU AI Act, and says it will provide tools for checking for its marks.

No, and be sceptical of any tool that claims it does. The text watermark is statistical. It lives in which words the model chose, so deleting invisible characters does not touch it. What this tool removes is a different and real problem: the invisible Unicode that turns up in AI output and quietly breaks spreadsheets, code, URL slugs, search and word counts. That is text hygiene, not evasion.

Not by any character-level tool. Reporting on comparable systems indicates the signal degrades under heavy paraphrasing, translation or full rewriting, because those change the word choices the mark is carried in. Short passages may also lack enough text to score reliably. This site does not provide a paraphraser and does not recommend one. If your goal is to pass work off as your own, no tool here will help with that, and the honest route is to write it.

Possibly, and this is the most under-discussed part. Because the mark is applied to text the model produces, asking Claude to proofread, translate or restructure your own writing can return copy that carries it. The EU AI Act carves out AI used as an assistive function for standard editing that does not substantially alter the content, but that is a legal distinction, not something the watermark itself can detect. A detected mark is not evidence that a human did not write the underlying work.

No. Commercial AI detectors work on statistical properties of word choice and sentence structure, not on invisible characters. Removing hidden Unicode changes nothing about how a detector scores the prose itself. Anyone selling a character stripper as a detector bypass is overselling it.

No. The entire tool is JavaScript running in your browser. There is no server component, no analytics call carrying your content, no logging and no storage. You can disconnect from the internet after the page loads and it will keep working, which is the easiest way to prove it to yourself.

U+200B is a character with no visible width. It occupies a position in the string, counts toward the character count, and shows nothing on screen. It has legitimate uses, such as marking line-break opportunities in long strings. But because it is invisible and survives copying, it is also the simplest way to mark a piece of text so it can be recognised later.

Unicode tag characters and variation selectors can encode ordinary bytes. A run of them sitting after a normal word is not random noise. It is data. When the tool finds such a run, it decodes it and shows you the message. This is steganography, and it is a real technique used to smuggle instructions into text that a person reads as clean.

Because they are the most recognisable stylistic signature of current language models, and many people want them gone for that reason alone. It is a style choice, not a correctness one, so it is a toggle. Turn it off if you write em dashes yourself and want them kept.

Invisible characters carry no meaning, so removing them is lossless. The typography options do change your text: a curly apostrophe becomes a straight one, an em dash becomes a hyphen. Those are off or adjustable by default. Use "Show what was removed" to see every single change before you copy.