Key features to distinguish between human- and AI-generated texts: What university professors say

Detection of Artificial Intelligence (AI)-generated texts is crucial in education for several reasons. This study provides direct evidence from university professors’ experiences regarding the key features they use to identify AI-generated writing and ranks these features based on their perceived importance. The research was conducted in two phases. In the first phase, online interviews were used to extract the most salient features that help professors detect AI-generated texts. In the second phase, an online survey asked professors to rate the likelihood that each identified feature contributes to successful AI text detection. The findings revealed that professors consider the absence of language errors, complex syntax and phrasing, use of difficult vocabulary, presence of hallucinated content, fabrication of nonexistent sources, repetition of words or phrases, and distinctive text structure as potential indicators of AI-generated text. Among these, hallucinations, fabrication of sources, and the absence of language errors were rated as the strongest cues for detection, while difficult vocabulary was seen as the least reliable indicator. These results have important pedagogical implications since they can inform the development of more effective detection tools, and guide the design of academic integrity policies and instructional strategies to address the challenges posed by AI-generated content.