Does ChatGPT put a watermark in text?
No published ChatGPT feature hides a watermark inside the text it writes. OpenAI has said it researched text watermarking but has not released it as a product. When people find “a watermark” in ChatGPT text, it is almost always one of the technical traces below — and those are real, visible to software and easy to check.
What the checker actually finds
| Trace | Example | Where it comes from |
|---|---|---|
| Invisible characters | U+200B zero-width space, U+202F narrow no-break space | Copying from a chat window or a web page |
| Link tracking tags | ?utm_source=chatgpt.com | Links that ChatGPT search cites |
| Citation leftovers | :contentReference[oaicite:0]{index=0}, 【4:0†source】 | Answers copied with sources |
| Hidden Unicode tags | U+E0041… | Text that hides instructions for another AI |
| Markdown | bold, ## headings | Chat formatting pasted as plain text |
| AI-style typography | word—word, curly quotes in the wrong style | Model output that ignores your language’s norms |
Each finding is shown in your text with its code. Nothing is guessed: if a character isn’t there, the checker doesn’t report it.
Watermarks a character scanner can’t see
Some providers use statistical watermarks: the model slightly prefers certain words, and a detector that knows the key can measure that pattern. Google’s SynthID Text works this way for Gemini. A statistical watermark lives in word choice, not in characters, so no character scanner — including this one — can find or remove it. Rewriting the text yourself is the only thing that changes it.
Other models
- Claude, Gemini, Copilot, DeepSeek and Perplexity answers pick up the same copy traces: special spaces, Markdown and, for tools with web search, citation markers and tracking parameters.
How to use the result
- Paste the text above. The check runs on your device.
- Remove invisible characters for free, or remove every trace with Pro.
- Copy the result for Word, Google Docs or email — or check a Word file on the document cleaner.
Remove AI Traces removes technical traces. It doesn’t rewrite your text or promise any AI-detector score. How each check works is described in the methodology.
Updated 2026-11-03