How discourse concentration and content valence drive engagement with conversational content

Conversational content, that is, content units organized around spoken interaction between two or more speakers, has become a prominent and economically significant form of communication across contemporary media environments, with podcasts alone reaching over 100 million monthly listeners and generating over $2 billion annual advertising revenue in the U.S. Despite the rapid growth of conversational media and formats such as podcasts and social media livestreams, little is known about what drives audience engagement with this type of content. Drawing on conversation analysis, this research demonstrates that discourse concentration, defined as the extent to which speaker contributions are concentrated among a few dominant voices versus distributed more evenly across many, interacts with content valence to shape audience engagement. We find that for positively-valenced content, a more concentrated conversational structure leads to higher engagement, whereas for negatively-valenced content, a more fragmented structure elicits higher engagement. These effects are mediated by empathy toward the speakers. We test this framework across two controlled experiments using transcript-based stimuli (Study 1) and AI-generated voice stimuli (Study 2), respectively, and the econometric analysis of 401 real podcast episodes comprising more than 42,000 speaker turns and approximately 209 hours of audio (Study 3). Together, these studies provide causal evidence for the interaction, progressively extending it from text-based to audio stimuli, and demonstrate its real-world relevance. These findings show that the same conversational structure can either enhance or reduce engagement, offering actionable insights for podcast producers, media platforms, and advertisers.