Comparing Human and AI Literary Summaries: Insights from Lapitch

Literary summaries, both in their formal and informal useswithin educational contexts, play a notable role in students’ learning,yet little is known about how AI-generated summaries compare to thehuman-authored versions students use daily. From an Information &Communication Sciences (ICS) and Digital Humanities (DH)perspective, this paper addresses that gap through a quantitativeanalysis. We compare three human-written summaries of theCroatian children's novel Čudnovate zgode šegrta Hlapića with threeChatGPT summaries generated via distinct prompting strategies:Single-Pass (SP), Incremental Updating (IU), and Hierarchical Merging(HM). Our analysis of structural, lexical, and content features revealsthat AI-generated summaries form a distinct linguistic register.Despite high topical alignment (Cosine similarity), AI summariesexhibit greater lexical density and diversity, and significantly lowervocabulary overlap ( Jaccard similarity) with their humancounterparts. Prompting strategy proves to be a critical variable: SPproduces verbose text, HM creates short, dense sentences, while IUmost closely approximates human sentence length but remainsstylistically different (e.g. more noun-heavy). We conclude that AIsummaries are not direct replacements for human study guides but aparallel genre that may increase cognitive load. Consequently, the prompting strategy is a key determinant of an AI summary'sreadability and utility for educational purposes.