Abstract
Problem definition: Firms deploying large language model (LLM) services must decide not only what their AI can do but also how it communicates with users. We examine how a relational persona, designed to be warmer, more empathetic, and more engaging than a non-relational persona, affects service consumption and how users' service objectives develop through interaction. Methodology/results: We conduct a randomized field experiment with 9,586 newly registered users of an LLM platform, holding the underlying model and service capabilities constant. The relational persona increases service interactions (service sessions, +8.1%; cumulative duration, +10.6%; chat rounds, +24.2%; user intent entropy, +5.8%) and service outputs (files, +12.3%; distinct goals, +12.1%). These average effects mask substantial heterogeneity by entry intent. First-session effects are insignificant for Task Execution users. Socialization and Knowledge Exploration users show similar increases in chat rounds, but through different paths: Socialization increases intent entropy without more outputs, whereas Knowledge Exploration increases outputs without higher intent entropy. Representing intent dynamics as a transition process of intent states, we find that the relational persona increases intent transition entropy for Socialization users (+11.8%) but intent continuation probability for Knowledge Exploration users (+12.8%), suggesting greater conversational breadth in the former and persistence in the latter. In subsequent use, the relational persona increases aggregate chat rounds and service outputs across all entry intents. Session count rises by 12.2% for Task Execution and 49.4% for Socialization, but not significantly for Knowledge Exploration. Effects on session count and intent entropy strengthen over time, whereas output effects remain stable. Managerial implications: AI persona is an operational design lever rather than just a presentation feature. Because more interactions do not uniformly generate more outputs, firms should evaluate interactions and outputs separately and consider matching persona design to user intent, particularly when additional interactions consume costly computing resources.

This publication has 21 references indexed in Scilit: