To Feel is to be Human? The Empirical Study of Reader Response to Original and AI-Generated Texts

This study follows the trend of examining real readers’ reactions to poetry and prose, but innovates by involving large language models to imitate the individual author’s style. The results of two experiments, in which respondents read either Poe’s “Annabel Lee” and “Morella” or their AI-generated versions, with the literary form being changed, are reported. Reactions to each text were probed, using Likert scales and semantic differential scales. The results point to statistically significant differences between the groups. The findings also indicate that Poe’s idiolect, idiostyle, and narrative techniques are perceived beyond the literary form of the texts. Surprisingly, readers confused the original texts with AI and vice versa, attributing AI stylistic features that characterise Poe’s style. This raises concerns about the distinction between human authorship and AI writings, becoming a prospective direction for future research.