Human, AI, and collaborative writing in computer science: a multilayer linguistic analysis of university students’ prompt-driven text production

This study comparatively examines three different production formats-human writing, text generated solely by an artificial intelligence (AI), and AI-human collaborative writing, in writing and summarization tasks under four different prompt strategies. Furthermore, four different prompt strategies shaping the production process were used to evaluate text quality, semantic consistency, and how discourse structure responded to these strategies. In the study, 81 university students performed writing tasks structured according to prompt leveling both as humans, with the Gemini, and in a hybrid production format. All generated texts were evaluated by students using a five-dimensional rubric and were also examined using multidimensional natural language processing (NLP) analyses (lexical diversity, syntactic complexity, semantic consistency, readability, discourse structure, and topic modeling). The findings show that human productions offer the highest lexical diversity, morphological richness, and discourse complexity, while AI productions provide high fluency, low semantic drift, and high local consistency. Collaborative productions created a balanced structure between the two extremes, combining both human contextual flexibility and large language model (LLM) structural fluency. Student ratings revealed that Gemini received highest scores in areas such as content quality and level of detail, while hybrid texts presented a strong intermediate profile in terms of overall satisfaction and originality.