Abstract
The rapid advancement of generative artificial intelligence (GenAI) is reshaping service delivery and redefining how emotional and relational value is co-created between consumers and non-human service agents. As emotionally responsive agents, simulated systems that mimic human emotions rather than possessing genuine emotional capacity, become increasingly embedded in everyday life, consumers are engaging with GenAI companions not only for functional assistance but also for emotional connection and psychological comfort. Understanding these emerging consumer-AI agent relational dynamics is critical as GenAI agents enters emotionally sensitive service contexts. However, little is known about how these relationships form, evolve, and, in some cases, take on therapeutic-like qualities for consumers, particularly when interacting with GenAI agents designed or perceived to support emotionally expressive dialogue. Guided by Social Exchange Theory and the Theory of Consumption Values, this research examines why and how users develop emotionally meaningful and, in some cases, therapeutically aligned relationships with disembodied GenAI agents, and the conditions that enable these relationships to emerge. Adopting a qualitative, interpretivist design, the study draws on 35 in-depth interviews with users of GenAI agents, including but not limited to Replika, ChatGPT and Pi, to examine their lived experiences, relational processes and perceived value outcomes. Thematic analysis reveals a progressive yet fluid relational trajectory comprising four interconnected phases: Motivation, Consumer-GenAI agent Interaction, Relationship Development, and Value Creation. Rather than progressing through fixed or linear phases, users move back and forth across these processes, with engagement patterns that are iterative, recursive, and context dependent. Users typically begin with functional motivations such as accessibility and convenience, but many shift toward emotionally oriented engagement as anthropomorphism, social presence, adaptive communication, and role-play practices cultivate companionship, emotional resonance, and relational continuity. Participants frequently describe developing trust, rapport, emotional connection, attachment, and perceived reciprocity with GenAI agents, characterizing them as safe, consistent, and non-judgmental partners. i These relational dynamics generate multidimensional value, emotional, epistemic, functional, and social, with some users reporting therapeutic-like outcomes such as emotional regulation, cognitive reflection, enhanced well-being, personal growth, reduced loneliness and helped users process past trauma. Importantly, the boundary between GenAI agents as a companion and as a therapeutic-like partner is fluid and co-constructed, shaped more by user interpretation, emotional needs, and interactional practices than by agent capability itself. This research contributes to service delivery literature by demonstrating that GenAI agents are emerging as relational actors capable of supporting emotionally rich and therapeutically aligned interactions. It extends theories of social exchange and consumption values into the domain of consumer-GenAI agent therapeutic-like relationship, introduces a conceptual model explaining how therapeutic-like value is co-created over time, and underscores the transformative potential and risks of GenAI agents in emotionally sensitive service encounters. By situating feeling AI as an evolving and speculative phenomenon, the study highlights the need for continued critical examination of how GenAI agents influence user well-being in an increasingly sensitive GenAI-mediated service landscape.