What’s different for you specifically
OpenAI’s watermark applies to ChatGPT and Codex for eligible EU users on all plans, rolling out over the coming weeks. Your colleague in another country doesn’t get it by default. If you work across borders, your draft may carry it and theirs may not, which is one more reason ‘it passed a check’ never settles anything on its own.
The mark sits in word choice, so it travels in copied text. OpenAI also says it can’t tell how much human editing or judgement went in. In other words, a mark describes the tool, not your contribution.
A sensible routine
Draft with whatever help you like, then read the result as your own editor. Cut the filler transitions, put in the detail only you have, and reorder anything that isn’t how you’d argue it. Keep your version history; it’s the best evidence of your process.
Then run the draft through NeverPrompted. It flags stock phrasing and flat rhythm, offers rewrites you approve one by one, and reports the share of your original words that changed. If a draft needs to move further, set a word-change target and see whether it can get there without losing meaning or facts.
Privacy, which matters more here
EU writers have good reason to dislike pasting client or student material into another upload box. NeverPrompted runs the check and the rewrite on your device and doesn’t send your text anywhere.
Wherever this page describes a result: a detected mark is not proof of authorship, and an absent mark is not proof of human authorship. NeverPrompted's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.