JMIR Publications

ABSTRACT

People increasingly consult generative artificial intelligence (AI) about symptoms, test results, emotional concerns, and whether professional care is needed. Existing research has largely examined these systems as information sources, relational interfaces, or agents entering the patient-clinician encounter. These perspectives do not fully explain how AI enters the earlier and wider social process through which illness is interpreted and pathways to care are constructed. We propose an extension to the Network Episode Model, the digital health alter: a temporary relational position enacted when a conversational AI system is selectively activated within a health episode and contributes to interpreting, communicating, emotionally processing, preparing for, or navigating the problem. Alter status is distinct from influence, which varies with perceived authority, reliance, and person-task-system-network fit. The framework shifts the unit of analysis from the AI output, interface, or clinical dyad to the entire health-seeking episode, including activation sequence, changing roles, network reconfiguration, and the visibility of prior AI influence. We integrate six literature streams, from health sociology and information seeking to trust, automation, and human-AI collaboration. The framework identifies six roles that AI may enact within a health episode and explains how their influence depends on the user, the task, the system, and the surrounding network of human and professional support. It also describes a hypothesized hidden influence pathway in which an earlier, undisclosed AI conversation shapes a later consultation without being visible to the clinician. Six testable proposition families and an empirical agenda translate the concept into measurable constructs. The framework is intended to move patient-facing AI research beyond answer accuracy toward the study of how human-AI interaction may reconfigure health networks, authority, communication, and care-seeking