Does ChatGPT watermark its text?
No. OpenAI built a text watermarking system and decided not to deploy it.
By OpenAI's own account the method worked well against local edits such as paraphrasing, but was defeated by translation or rewording with another model, and applying it at ChatGPT's scale would produce a large absolute number of false positives even at a low rate. They chose cryptographically signed metadata instead, which cannot produce a false positive, and SynthID for audio. None of that touches the text you copy out of a chat.
What Clipboard Sanitizer can do about it
Quite a lot, because everything ChatGPT leaves in your text is a character rather than a signature. The narrow no-break space is the well-documented one, reported in GPT-5 output in place of ordinary spaces. Zero-width characters and typographic punctuation turn up in the ordinary way. All of it is removable, and all of it is what this app removes.
Questions
Why do people say ChatGPT watermarks its output?
Because invisible characters do turn up in it, and a pattern of invisible characters can be used to identify machine-generated text whether or not anybody intended it. That is a real thing to care about. It is just not a watermark, and OpenAI has not described it as one.
Did our own sample contain anything?
No, and that proves very little: the same paste path had already flattened this sample's dashes. See the caveat above.