Experiment · Core reading
Detection of GPT-4 Generated Text in Higher Education: Combining Academic Judgement and Software to Identify Generative AI Tool Misuse
Paper at the publisher Free to read
In short
The authors report that an AI-writing detector flagged 91% of their 22 ChatGPT-written test submissions, but only 54.8% of the AI text in them; the 15 staff marking them referred 54.5% for misconduct.
What it looked at
22 experimental submissions produced with ChatGPT, using prompts meant to make AI flagging less likely. 15 academic staff marked them alongside genuine student submissions, with an AI-writing tool available. The authors compared what was flagged, what staff referred for misconduct, and the grades awarded.
What it means for teaching
May inform thinking on assessment design and on how staff read detector reports, from one experiment with 22 submissions and 15 markers.
Limits
One detector, 22 test submissions and 15 markers. The AI-written work was prompted to make flagging less likely, so the figures may not reflect how students ordinarily use such tools.
Cited 382 times by other papers, as counted by OpenAlex in November 2026. A count measures attention, not quality.