For centuries, scholars, critics, and authors have wrestled with a fundamental query: What makes a story a “good” story? (Chatman 1978; Li 1996) From Roland Barthes’ structural analysis of narrative in the 1900s to the Reddit users asking the internet “What makes a good story?” in 2018, the humanities have long sought to decode the mechanics of fiction. Writers like William Zinsser have defined the craft through clarity and humanity, while researchers like Paul J. Zak have explored the neurobiological basis for why the brain craves narrative arcs (2013; 2014). However, the public release and rapid adoption of Large Language Models (LLMs) such as ChatGPT, Claude, and Gemini have brought these age-old questions back to the forefront of discussion in a radically new context. When an algorithm can generate a novel-length manuscript with the touch of a button, the definitions of authorship, creativity, and intelligence become increasingly murky, forcing us to ask whether these terms are enough to prove our humanity (Hajibayova / Lee 2025; Marchant 2025).

The implications of this shift are not merely theoretical, but also practical and pervasive. Artificial intelligence is no longer a fringe experiment but a tool increasingly integrated into the creative workflow. As noted by Fang et al., there is a significant rise in the systematic use of AI technologies for story writing, fundamentally altering the landscape of literary production (2023). Consequently, determining what distinguishes a human writer from an artificial one is no longer a philosophical exercise but an urgent necessity for the future of the arts and humanities.

To answer these questions, an interdisciplinary approach is crucial. While the humanities provide the framework for understanding the essence of a writer—often linked to self-regulation and intent—technical tools are required to empirically validate these claims and analyze how AI models operate within the narrative domain (Ferrari et al. 1998). Recent computational research in this arena has largely focused on Human-Computer Interaction (HCI) or stylometric comparisons, analyzing the linguistic fingerprints and stylistic choices of human versus AI authors (O’Sullivan 2025; Gunser 2021). While valuable, these studies often analyze how a story is told(the syntax and vocabulary) rather than what is told(the structural components of the narrative itself).

To address this gap, our research pivots from stylistic analysis to an ontology-driven examination of story elements. We argue that to truly understand the storytelling capabilities of modern LLMs instead of just their ability to mimic good style, we must look beyond surface-level text generation and quantify the fundamental building blocks of their narratives: the characters, events, and relationships. Ontologies provide a standardized, machine-readable method for this type of structured representation (Blin et al. 2025). While various frameworks exist, such as the Storytelling Ontology Model or the Drammar Ontology, we utilize the GOLEM Ontology for Narrative and Fiction (Pianzola et al. 2025). This specific ontology allows for a granular mapping of narrative components, enabling a direct quantitative comparison between human and machine creativity.

This study presents a comparative study of human and AI storytelling practices designed to isolate narrative competence. We engaged human participants and three distinct free AI models (ChatGPT 5.1, Claude Sonnet 4.5, and Gemini 2.5 Flash Lite) in a controlled writing task. Each group was presented with a “subpar” story synopsis in the form of a film treatment deliberately lacking in polish and were given a 30-minute window to rework it to their liking. The prompt challenged them to exercise creative license, allowing changes to characters, conflict types, or events to transform the synopsis into what they perceived as a “well-written story.” This process was repeated 34 times within each group resulting in 136 revised story synopses.

By holding the initial premise and time constraints constant, we successfully isolated the variable of narrative instinct. We then applied the GOLEM ontology to the resulting stories, mapping the density and distribution of specific elements such as social relationships, character introductions, settings, and inference. This methodology moves beyond subjective quality assessment, using quantitative methods as a preliminary search for structural patterns. Our analysis aims to reveal whether human writers prioritize different narrative elements—such as complex social relationships—compared to the generative patterns of AI models. By combining the qualitative depth of literary theory with the quantitative rigor of knowledge graphs, this study offers a novel perspective on the evolving definition of authorship in the age of artificial intelligence.