The quick answers
Is it visible? No. It’s a pattern in which words the model chose, not a symbol or hidden file data.
Does it survive copy and paste? Yes, because it lives in the words themselves.
Does it identify me? OpenAI says no. It can indicate that an OpenAI system generated or processed part of a passage, not who asked for it.
Can anyone check it? Right now only approved researchers and expert organisations have access to OpenAI’s detector, so ordinary users and most institutions can’t.
Why only the EU?
The EU AI Act’s transparency rules, in effect since 2 August 2026, require AI-generated content to be marked in a way other systems can identify. OpenAI is meeting that where it applies and has said it isn’t making text watermarking a global default at launch.
Anthropic took the other route in August and said it will watermark Claude’s text everywhere.
What happens if I edit it?
The mark weakens. OpenAI’s own test showed detection falling from about 92% to 66% when 10% of words were replaced with synonyms, and short passages, math answers and translated text are harder to detect.
That doesn’t make a clean text proof you wrote it. OpenAI says a missing watermark doesn’t prove human authorship, because the text could be too short, heavily edited, or from a different AI.
What NeverPrompted does about it
NeverPrompted can’t read OpenAI’s watermark, because nobody outside OpenAI holds the key, and it says so on every result. What it does is help you rewrite your own draft so it genuinely sounds like you, entirely on your device, and show you what percentage of your original words changed along the way.
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.