Free tool, runs in your browser
AI Watermark Remover and Text Cleaner
Paste text from ChatGPT, Claude or Gemini. Write It strips the hidden characters, citation leftovers and formatting fingerprints AI chat apps leave behind, then tells you straight what a cleaner cannot remove.
- Nothing uploaded
- Cleans as you type
- Free, no sign-up
Watermark check
- Hidden-character marks
- Paste text to check
- AI citation and link tags
- —
- Formatting fingerprints
- —
- Statistical watermark
- Claude (models since August 2026), ChatGPT in the EU (from October 2026) and Gemini mark text through word choice. It cannot be seen, and cleaning characters does not change it. Rewriting the wording does.
Cleaning options
Invisible characters
- Hidden characters
Zero-width spaces and joiners, byte order marks, soft hyphens, direction marks, tag characters.
- Special spaces
No-break, narrow no-break and other odd spaces become normal spaces.
AI leftovers
- AI citations and link tags
ChatGPT citation leftovers and utm_source=chatgpt.com on links.
- Numbered citations
Markers like [1] or [2, 3]. Off by default, since essays use them too.
Formatting fingerprints
- Markdown
Asterisks, # headings, tables, code fences and LaTeX math.
- Em and en dashes
Replaced with a hyphen or a comma.
- Curly quotes
“ ” ‘ ’ become straight quotes.
- Ellipsis character
The single … character becomes three dots.
- Spacing
Double spaces, trailing spaces and extra blank lines.
Advanced
- Fancy letters
Fullwidth, ligature and bold-math letters become plain letters.
- Lookalike letters
Cyrillic and Greek letters hiding inside English words.
- Emoji
Removes emoji such as 🚀 and ✅.
What an AI watermark actually is in 2026
People use the word for two very different things, and the difference decides what any tool can do about it.
- Character marks. Invisible Unicode characters inside the text: zero-width spaces, no-break spaces, joiners and the like. They are what most "ChatGPT watermark" posts are about. Whether they arrived on purpose or by accident, they are real characters, so they can be found and deleted. This tool does that.
- Statistical watermarks. A pattern in which words the model picked. Anthropic marks text from Claude models released since August 2, 2026, worldwide. OpenAI announced on October 5, 2026 that ChatGPT and Codex text in the EU will carry one. Google has marked Gemini text with SynthID since 2024. There is no character to delete, because the mark is the wording itself.
What this tool removes
| What it finds | Examples | What happens |
|---|---|---|
| Zero-width characters | U+200B, U+200C, U+200D, U+2060, U+FEFF | Removed. Joiners inside emoji and in scripts like Persian and Hindi are kept. |
| Special spaces | No-break U+00A0, narrow no-break U+202F, thin and hair spaces | Turned into normal spaces |
| Direction and tag characters | U+202E, U+2066, U+E0000 to U+E007F | Removed, and any hidden text they spell out is shown to you |
| ChatGPT citation leftovers | :contentReference[oaicite:0]{index=0}, citeturn0search3, 【4†source】 | Removed |
| AI tracking tags on links | ?utm_source=chatgpt.com | Removed; the rest of the link still works |
| Formatting fingerprints | Em dashes, curly quotes, the … character | Replaced with plain keyboard punctuation |
| Markdown | **bold**, ## headings, tables, \( LaTeX \) | Turned into plain text, or copied as real formatting for Docs |
Can any tool remove the Claude or ChatGPT watermark?
Not by cleaning. A statistical watermark survives copy and paste, retyping and file conversion because it lives in the words. Character removers, this one included, leave it exactly where it was. Tools that promise to strip it in one click are removing characters, not the watermark.
What weakens it is changing the words. OpenAI says that in one test, replacing 10% of words with synonyms dropped detection from about 92% to 66%, and that short passages and translated text are harder to detect. Anthropic says its mark may survive some editing. So the honest next step after cleaning is a real rewrite, in your own words or with the Write It Humanizer. We do not promise it defeats any watermark or detector: nobody outside Anthropic, OpenAI and Google can test against their keys.
Who can check for a watermark today?
As of October 5, 2026, Anthropic has not released a public text detector, and OpenAI is giving detector access only to approved researchers and expert organizations. Google offers SynthID detection for content made with its own models.
A watermark is not proof of cheating on its own, either. People use Claude and ChatGPT to edit and translate their own writing, and short text is hard to judge. For the full story, read what the Claude and ChatGPT watermarks mean.
How to clean AI text
Paste your text.
Copy from ChatGPT, Claude, Gemini, a PDF or a web page and paste it into the box. It is cleaned as you type.
Check what changed.
Open What changed to see every hidden character tagged where it was, plus the dashes, quotes and markdown that were fixed.
Pick your options.
Keep em dashes or turn them into commas, keep link URLs, or remove only the invisible characters.
Copy it out.
Copy plain text, or use Copy for Google Docs to paste headings, bold and lists as real formatting instead of symbols.
Writing in Google Docs?
Write It types, humanizes and checks your writing inside Google Docs, Slides and Word. Install the extension to try it free.