“Talk to Me, Don’t Scroll”: Voice-Based AI Interventions to Curb Social Media Overuse

In this TREO session, we discuss the potential for AI-powered, voice-based conversational agents to interrupt maladaptive digital behaviors such as compulsive social media scrolling. While the role of AI in marketing and personalization has been extensively studied (Huang & Rust, 2021), its application in promoting digital self-regulation and well-being remains underexplored. This conversation is timely, as platforms continue to optimize for engagement, even at the cost of user health (Berthon et al., 2019). Drawing on the Elaboration Likelihood Model (Petty & Cacioppo, 1986) and Feedback Intervention Theory (Kluger & DeNisi, 1996), we explore how chatbot voice pitch and feedback framing (corrective, outcome-based, or personal) can influence users' intention to continue or pause social media use. We also position voice-based feedback as a cognitive resource-preserving mechanism, in line with Conservation of Resources Theory (Hobfoll, 2001), offering an alternative to traditional self-control–based interventions. This talk raises important questions about the ethics and effectiveness of "well-being by design" approaches, where technology that contributes to overuse is also repurposed to mitigate it. Based on an ongoing experimental study involving AI voice interventions, we propose that personalized voice prompts from human-like chatbots may trigger deeper reflection and disengagement by activating the central route of persuasion, particularly among high-SMA users. Some questions we try to discuss in this session are: 1. How can we design persuasive AI chatbots that intervene in harmful digital habits without undermining user autonomy or trust? 2. Should social media platforms bear responsibility for embedding such interventions, or does this open the door to digital paternalism?