Remove AI watermark from text

Paste your text and the invisible clutter comes out: the zero-width spaces, tag characters and odd spaces that survive a copy-paste and quietly break your CMS, your code and your forms. What's still doing a spelling job, like Persian's joining marks or an emoji's joiners, stays. Your words are left exactly as you wrote them. Free, no sign-up, nothing uploaded.

0 files processed across our free tools
and counting

also normalise smart quotes & dashes

Cleaning happens in your browser: that text never leaves this page, and there is no upload, no account and no limit. Rewrite is the exception: that one sends your text to our server.

Works with any assistant, because it matches characters

ChatGPT, Claude, Gemini, Copilot, Grok, DeepSeek, Perplexity, Llama: the tool never asks where your text came from. It looks for characters that should not be in running text and takes those out. That is also why it works on text pasted out of Word, Google Docs, a PDF or a web page, which carry exactly the same debris.

Also known as an AI text cleaner

AI text cleaner is a fair name for this page: cleaning happens in your browser, with no upload, no account and no limit. It also works as an AI watermark remover for the invisible kind of watermark, and what comes out is Clean AI text: the same words, minus the invisible characters. If you want to Remove hidden characters online without installing anything, this is the plainest way to do it, and AI text cleanup takes one click. Some problems have a page of their own. If you want a text cleaner, or a way to clean text before it goes into a form, a spreadsheet or a CMS, those pages start there. If a single hidden character is breaking an import, that page shows you how to find it and remove it in one pass. And the ai text remover page covers the same job from the AI side, including what no cleaner can do. There is one tool on this page and it is free either way.

What it takes out

Zero-width characters. Zero-width space, joiner, non-joiner and word joiner: invisible on screen, but they split words as far as your CMS and your search are concerned.
Unicode tag characters (U+E0000: U+E007F). A block with no visible form at all, which is exactly why it is used to hide data inside plain text. Most cleaners skip it.
Direction marks and isolates. Left-to-right and right-to-left overrides that can make text display in an order it is not actually written in.
Variation selectors, the byte-order mark, soft hyphens and private-use characters.
Odd spaces, non-breaking, thin, hair, en, em and ideographic, become one normal space.
Optionally, curly quotes and dashes become straight quotes and hyphens.

characters are detected by name, plus the three private-use ranges. Want the longer version: what each of these is, and where it typically comes from (a chat assistant, a PDF, Word)? Invisible characters, explained. If you think of this as Unicode housekeeping rather than watermark removal, the unicode cleaner page explains the identical list from a developer angle. The special characters remover page is explicit about which ones this does and doesn't touch. Either way it's the same invisible character remover underneath.

Per assistant

AI watermark checker, want to see what's hiding in a piece of text before you decide to clean it? That page only detects, it never changes anything. Want to change how your text looks instead? The font generator turns it into bold, cursive, small caps and a dozen other Unicode styles: same browser-only approach, different job. And to turn your own handwriting into a real font you can install, use handwriting to font: print the template, snap a photo, download an OTF file.

Questions

Does this remove the AI watermark from my text?

It removes the invisible characters, zero-width spaces, joiners, tag characters and the like, that get carried along when you copy text out of an AI assistant. It does not touch your words. Statistical watermarks such as SynthID live in the word choice itself, so no cleaner can remove those without rewriting your text - and the cleaner rewrites nothing. There is a separate Rewrite button for people who want their wording changed; that is a writing aid, not a way around detection.

Which AI assistants does it work with?

All of them. The tool matches characters, not sources, so text from ChatGPT, Claude, Gemini, Copilot, Grok, DeepSeek, Perplexity or Llama is handled identically. It catches the same characters when they come from Word, Google Docs, a PDF or a web page.

How many characters does it detect?

named characters, plus the three private-use ranges: including the Unicode tag range U+E0000 to U+E007F, the block used to hide data inside plain text, which contain no standard characters at all and which most cleaners miss entirely.

Is a text watermark the same as an invisible character?

No. A statistical text watermark is a pattern in word choices, while an invisible character is a code point with no visible shape, like a zero-width space. This tool finds and names the second kind only.

Is my text uploaded anywhere?

The cleaner runs entirely in your browser: that text never leaves the page, there is no account, and there is no limit on how much you clean. The separate Rewrite button is the one exception: it sends your text to our server, which passes it to Anthropic, whose AI model rewrites it. Only Anthropic receives it, to rewrite it, and Anthropic does not train its models on it. We keep no copy of your text or the rewrite.

Why does my text break my CMS or code editor?

Invisible characters survive a copy-paste but are not what they look like. A zero-width space inside a variable name breaks the code, a non-breaking space breaks a search, and a smart quote breaks a string literal. Nothing looks wrong on screen, which is what makes it hard to find.

What this tool cannot remove

There are two different things people call an "AI watermark".

Invisible and typographic characters. This tool removes named characters that carry no visible glyph at all, as defined by the Unicode standard: not a number we made up. It also turns different space variants, including the narrow no-break space, into an ordinary space. Curly quotes and dashes go through a separate, optional step, typographic replacements, switched on by default; turn that option off and they stay exactly as typed. It also clears the three Unicode private-use ranges, including the tag range U+E0000, U+E007F used to hide data in plain text, those ranges hold no standard characters of their own, so we describe them rather than dress them up as a character count.
A statistical watermark in the word choice itself. Google's SynthID, which Google says has marked text in the Gemini app and on the web since May 2024, steers which words get picked and adds no characters at all, so no character cleaner, ours included, has anything to catch. Anthropic says text from its newer Claude models (launched on or after 2 August 2026) carries a statistical watermark applied as the model picks its words, with nothing added to the text (anthropic.com), so no character cleaner touches it. OpenAI announced one for ChatGPT in October 2026 (as reported by the press) that is statistical too, starting in the EU. The full picture: what a statistical watermark actually is, who has one live today, and how well it survives editing: is written up on its own page.

If a tool promises to strip "the AI watermark" without making this distinction, it is describing the first thing while letting you believe it does the second.

By Sam Ridder, I build and run TextNinja on my own hardware, on my own. Who I am.