Understanding User Navigation Behavior in Conversational Systems: An Empirical Study

Abstract

Despite rapid adoption of conversational interfaces in web and mobile applications, fundamental patterns of user navigation behavior within these systems remain poorly understood. In contrast to conventional graphical user interfaces where navigation is explicit, conversational systems often blur task boundaries. This study analyzed usage logs from a multilingual conversational assistant deployed across e-commerce and travel planning services over a six-month period. Aggregate data from ~98,000 sessions indicate that users employ implicit navigational cues, such as entity rephrasing and confirmatory follow-ups, at significantly higher rates than previously reported. Specifically, 39% of task completions required at least two context switches, with an average of 3.4 user–agent turns per navigation phase. Qualitative feedback revealed that users rely on conversational context retention more heavily on mobile devices than on desktops, suggesting modality-dependent design considerations. The findings motivate refinements in session modeling and adaptive context preservation mechanisms for future dialogue systems.