Large language models (LLMs) have introduced new challenges to academic integrity, particularly regarding the appropriation of AI-generated outputs as original human authorship and the difficulty of verifying independent work. While some universities and academic publishers increasingly require explicit disclosure of the use of artificial intelligence (AI), the scope and implementation of these requirements remain inconsistent. This paper examines current practices related to AI use, focusing on LLM-based ghostwriting and the reliability of disclosed interactions as evidence of authentic use. The study includes an experimental component involving AI-assisted essay generation, highlighting practical and ethical dilemmas associated with academic integrity. It further explores the possibility of mimicking authentic interactions, which raises concerns about the effectiveness of current approaches. To investigate these questions, a survey was conducted among teaching staff at the Faculty of Computer Science and Engineering (FCSE) in Skopje to assess their ability to identify AI-generated essays and their trust in disclosed interactions. Among the 28 respondents, a majority (82.14%) indicated that it is possible to identify AI-generated content based solely on language style, while 64.29% reported detecting linguistic inconsistencies that could result from the use of LLMs. Despite noticing AI-related linguistic markers, only 53.57% concluded that the essay was not human-written. This view was shared by just 27.27% of assistants, compared to 70.59% of professors, whose extensive experience appeared to help them recognize that a substantial portion of the text had been AI-generated. The findings are discussed in the context of teaching experience and existing policies, leading to recommendations for improving student assessment and strengthening the ethical use of generative artificial intelligence (GenAI).
Katerina Zdravkova (Mon,) studied this question.