Unofficial. Not affiliated with OpenAI. ChatGPT is a trademark of its owner.
ChatGPT watermark
textGrain is a statistical pattern in ChatGPT’s word choices. OpenAI announced an EU rollout for ChatGPT and Codex text, with API watermarking available on some models but off by default. A hidden-character scan will not show textGrain, because there is no invisible character to find.
Scan to list hidden characters. Your text stays in this tab.
What OpenAI announced
On 2026-10-05, reporting on OpenAI’s text provenance work described watermarking for EU ChatGPT and Codex text, optional API controls (default off), and a detector limited to approved research groups. Those details describe a word-choice signal, not zero-width Unicode.
Official rewrite test numbers
OpenAI’s own lab test (English ELI5, about 400 tokens) reported detection near 92% before edits, about 66% after swapping 10% of words, and about 17% after swapping 25%. That is their result, not a promise about your text.
Check this text for hidden characters anyway
Invisible Unicode can still appear when text is copied between apps. Use the scanner above, then see ChatGPT watermark remover and Pricing.
FAQ
Does ChatGPT still insert zero-width characters?
We have no evidence that textGrain uses them. Scan if you want to see invisible characters; a clean scan does not mean the text is human.
Will a synonym swap remove it?
OpenAI’s own test says a 10% swap cut detection from about 92% to 66%, and a 25% swap to 17%. That is their lab result, not a promise about your text.