AI Watermark Remover for Text
Works on output from any model. Paste text from ChatGPT, Claude, Gemini, Copilot or anything else and get back a copy with the hidden characters gone.
0 chars
Your text is processed in memory and never stored. See the full list of characters this removes.
The same cleaner, whatever wrote the text
Hidden characters are not a ChatGPT problem specifically. They come from the model, from the web page you copied through, from a word processor, or from a CMS that helpfully converted your quotes. The cleaner does not care about the source. It looks at the codepoints.
Which models actually watermark text
As of September 2026, this is where things stand. The distinction that matters is character versus statistical, because only one of them can be removed.
| Model | Statistical watermark? | Hidden characters? |
|---|---|---|
| ChatGPT / OpenAI | None shipped for text | Sometimes |
| Claude / Anthropic | Yes, SynthID-Text from 11 Aug 2026 | No, by design |
| Gemini / Google | Yes, SynthID since 2024 | Sometimes |
| Copilot, Llama, Mistral | Not announced for text | Sometimes |
Anthropic and Google both signed the EU Code of Practice on Transparency of AI-Generated Content; OpenAI signed it too but has not shipped a text watermark.
FAQ
Frequently asked questions
- Does this work on Claude output?
- Yes. It removes any hidden characters in Claude output, the same way it does for any other source.
- Does it work on Gemini output?
- Yes. Gemini output goes through exactly the same character rules as everything else.
- Is there a limit on how much text I can clean?
- The free web tool handles up to 20,000 characters at a time. The API allows 100,000 characters per request on Starter, 250,000 on Growth and 500,000 on Scale.
Last reviewed .