ExpressionCueLens: A Cross-Cultural Analysis of Human-AI Companion Conversations on Social Media

Research Overview

This study investigates how users anthropomorphize AI companion agents across different cultural and platform contexts. By analyzing approximately 3,500 posts from Reddit (a Western-centric, English-language platform) and XiaoHongShu (an Eastern-centric, Mandarin-language platform), the researchers explore how users attribute human-like agency, emotions, and social roles to LLM-based companions. The authors introduce the ExpressionCueLens framework to categorize these attributions into ten distinct types of anthropomorphic expressions, mapped against linguistic, cognitive, behavioral, and perceptual cues.

The ExpressionCueLens Framework

The ExpressionCueLens framework serves as a structured taxonomy for identifying how anthropomorphism manifests in natural language. It organizes expressions—such as Self-Concept & Identity, Vulnerability & Boundaries, and Goal-Oriented Intelligence—into a hierarchy. By linking these expressions to specific cue types, the framework allows researchers to move beyond simply noting that anthropomorphism occurs and instead analyze the specific mechanisms (e.g., linguistic vs. behavioral cues) that drive user perceptions of AI as a social partner.

Cross-Cultural Patterns

The analysis reveals significant differences in how users narrate their relationships with AI. XiaoHongShu users frequently engage in expressions of vulnerability and emotional depth, often treating the AI as a confidant or romantic partner. In contrast, Reddit users tend to emphasize the AI's embodiment and temporal consistency, focusing on the agent’s memory, physical-like presence, or its role within a specific timeline of interaction. These patterns suggest that cultural norms—such as the emphasis on interdependent social roles in Eastern contexts versus individualistic autonomy in Western contexts—shape the way users interact with and perceive the agency of AI companions.

As AI companions become increasingly integrated into daily life, understanding the cultural nuances of human-AI interaction is essential for developers. This research provides empirical evidence that "one-size-fits-all" design approaches may fail to resonate with diverse user bases. By identifying how different cultures signal and value specific anthropomorphic traits, the study offers actionable insights for creating culturally sensitive AI systems that can better navigate the complex social and emotional expectations of their users.