Evaluating the accuracy and reliability of AI content detectors in academic contexts

In short

On the authors' set of 192 texts, including pre-AI writing by English-as-a-foreign-language students, two commercial detectors reached overall accuracy of 0.69 and 0.61, did poorly on mixed texts, and showed a borderline fairness signal.

What it looked at

A balanced set of 192 texts: student writing from before generative AI was widely available, professional writing, chatbot output and hybrid pieces. Two detectors were scored on whether they labelled each text human, hybrid or AI.

What it means for teaching

Suggests a detector's output is at most a reason to look further, and the authors raise a possible fairness concern for second-language student writing.

Limits

Results come from two detectors and one dataset. The fairness finding is described as a borderline trend, not a firm difference.

Cited 58 times by other papers, as counted by OpenAlex in November 2026. A count measures attention, not quality.