The author vs AI: the empirical research of Poe’s poetics
Nika Marushchak · Borys Grinchenko Kyiv University Institutional repository (Borys Grinchenko Kyiv University) · 2025
The study offers an innovative approach to examining the perception of literary texts, employing empirical research methods that utilise large language models (LLMs). We report the findings of two experiments, in which the respondents read either Poe’s “Annabel Lee” and “Morella” or their AI-generated prose and poetry versions, respectively. The results indicate that Poe’s idiolect, idiostyle, and idionarration were recognised by readers regardless of the genre. Remarkably, the respondents often confused Poe’s original texts with those generated by AI. They attributed artificial intelligence stylistic features that are characteristic of Poe, which highlights concerns about the discrepancy between human authorship and AI-generated texts.