Pleasurable experiences of verbal poetry: Words speak louder than voices.
Alisha Isherwood, Andrea Piovesan, Ruby Isserlis, Michel Belyk
Psychology of Aesthetics Creativity and the Arts · 2026
Poetry is an enduring cultural form of expression that predates written language. Its distinctive use of poetic devices—such as rhyme, meter, repetition and metaphor—is thought to enhance its aesthetic impact. Yet the experience of poetry is mediated through performance, and performers vary in their ability to translate texts into compelling spoken word.
We leveraged recent advances in generative AI to produce poetic readings of contemporary human-composed poetry, AI-composed poetry, or non-poetic control texts, performed by a voice AI or a professional human actor, in a mode instructed/parameterised to be highly expressive or inexpressive. Listeners strongly enjoyed poems more than control texts and consistently preferred the AI-composed poems to human-composed poetry. In contrast, they displayed no preference for expressive over inexpressive performances (Experiments 1 and 2), despite strong indications that both AI and human performers made extensive use of vocal pitch and timing cues (Experiment 3).
Sentiment analysis revealed that AI-compositions included more consistent emotional sentiments that are easily decoded from individual lines of text with little need to integrate across the whole poem (Experiment 4). Combined with the tendency for AI-composed poems to adhere rigidly to traditional patterns of rhyme and meter relative to the free verse favoured by contemporary western poets, the findings highlight that traditional forms may drive perceptual fluency and thereby induce a greater sense of subjective enjoyment. Further research should leverage the evident palatability of AI-composed and AI-performed performances for systematic and controlled, but holistically meaningful, experiments on the reward value of poetic devices.