The voice or the script? Unlocking AI agent persuasion through modality and mental imagery

AI agents are increasingly used in marketing to deliver personalized product recommendations through text- or voice-based interactions. While prior research has focused on content characteristics, the influence of communication modality remains underexplored. This research investigates how the modality of AI agent communication (text vs. voice) shapes consumers’ recommendation adoption behavior and how this effect varies across message contexts. We first conducted a field experiment in collaboration with a local mobile phone retailer, which provided initial behavioral evidence that text-based AI recommendations lead to higher engagement than voice-based recommendations. Building on this real-world pattern, three controlled experimental studies further unpack the underlying psychological mechanism and boundary conditions. Across these studies, we show that text-based recommendations generally elicit stronger recommendation adoption behavior by facilitating vivid mental imagery. However, when embodied cognition cues are present, voice-based recommendations become more persuasive due to their alignment with sensory simulation. This pattern is further moderated by temporal framing: under near-future framing, voice with embodied cues evokes stronger imagery and engagement, whereas under distant-future framing, text remains more effective. Together, these findings position mental imagery as a core explanatory mechanism through which AI agent modality translates into consumer response, and demonstrate how aligning modality with sensory and temporal contexts can enhance the effectiveness of AI-mediated recommendations.