Paper Number

ICIS2025-2311

Paper Type

Short

Abstract

Artificial intelligence (AI) chatbots are increasingly used for social interaction, raising questions about how anthropomorphic design cues influence user engagement. We focus on a novel strategy: AI self-disclosure of social ties (e.g., referring to “my acquaintance,” “my friend,” or “my close friend”). Drawing on theories of self-disclosure, anthropomorphism, and the uncanny valley of mind, we designed a randomized field experiment on a commercial AI character chat platform in South Korea. Over several thousand user were randomly assigned to messages varying in tie-strength references. Preliminary findings suggest that self-disclosure of social ties increases conversational engagement, primarily by intensifying interaction once users respond. Stronger tie references did not consistently outperform moderate ones, and prior relationship depth did not systematically moderate effects. This pilot study demonstrates the feasibility of testing relational cues in real-world AI platforms and highlights plausibility as a boundary condition for anthropomorphic design.

Comments

15-Interaction

Download

Share

COinS
 
Dec 14th, 12:00 AM

AI with Friends?: The Effects of Chatbot Social Tie Self-Disclosure on User Engagement

Artificial intelligence (AI) chatbots are increasingly used for social interaction, raising questions about how anthropomorphic design cues influence user engagement. We focus on a novel strategy: AI self-disclosure of social ties (e.g., referring to “my acquaintance,” “my friend,” or “my close friend”). Drawing on theories of self-disclosure, anthropomorphism, and the uncanny valley of mind, we designed a randomized field experiment on a commercial AI character chat platform in South Korea. Over several thousand user were randomly assigned to messages varying in tie-strength references. Preliminary findings suggest that self-disclosure of social ties increases conversational engagement, primarily by intensifying interaction once users respond. Stronger tie references did not consistently outperform moderate ones, and prior relationship depth did not systematically moderate effects. This pilot study demonstrates the feasibility of testing relational cues in real-world AI platforms and highlights plausibility as a boundary condition for anthropomorphic design.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.