Teachers can check a student text for ChatGPT writing with a style-based detector such as ChatGPTDetect. They cannot use OpenAI's watermark detector, because OpenAI puts checking student work outside the scope of its program.
Why the watermark does not help in a classroom
OpenAI's textGrain watermark applies only to ChatGPT text in the EU, and its detector is restricted to approved researchers and expert organizations. Text from the rest of the world is not marked, and no mark does not mean a person wrote it. Details are on the ChatGPT watermark status page.
What a style-based check gives you
Paste the text and the checker reports whether it reads as AI-written, highlights the passages that carry the signal, and shows how close the style is to ChatGPT and other families. At least 100 words are needed. Measured on human-written text in seven languages, 0.5% was wrongly flagged as AI.
What to do with a result
- Read the highlighted passages yourself, not just the score.
- Compare with the student's earlier work in class.
- Ask the student to talk through the argument or the sources.
- Treat the result as one signal. OpenAI itself says detectors are not reliable enough for judgments about students, which is why a flag should start a conversation, not end one.
Schools and larger volumes
The free check covers one text. For repeated checks, files and records you can keep, the paid tier of the same engine covers them. For the wider field, see the best ChatGPT detectors, and for the phrases that give AI text away, how to tell if ChatGPT wrote it.