Humanlike AI Design Increases Anthropomorphism but Yields Divergent Outcomes on Engagement and Trust Globally

Abstract: Over a billion users across the globe interact with AI systems engineered with increasing sophistication to mimic human traits. This shift has triggered urgent debate regarding Anthropomorphism, the attribution of human characteristics to synthetic agents, and its potential to induce misplaced trust or emotional dependency. However, the causal link between more humanlike AI design and subsequent effects on engagement and trust has not been tested in realistic human-AI interactions with a global user pool. Prevailing safety frameworks continue to rely on theoretical assumptions derived from Western populations, overlooking the global diversity of AI users. Here, we address these gaps through two large-scale cross-national experiments (N=3,500) across 10 diverse nations, involving real-time and open-ended interactions with an AI system. We find that when evaluating an AI's human-likeness, users focus less on the kind of theoretical aspects often cited in policy (e.g., sentience or consciousness), but rather applied, interactional cues like conversation flow or understanding the user's perspective. We also experimentally demonstrate that humanlike design levers can causally increase anthropomorphism among users; however, we do not find that humanlike design universally increases behavioral measures for user engagement and trust, as previous theoretical work suggests. Instead, part of the connection between human-likeness and behavioral outcomes is fractured by culture: specific design choices that foster self-reported trust in AI-systems in some populations (e.g., Brazil) may trigger the opposite result in others (e.g., Japan). Our findings challenge prevailing narratives of inherent risk in humanlike AI design. Instead, we identify a nuanced, culturally mediated landscape of human-AI interaction, which demands that we move beyond a one-size-fits-all approach in AI governance.

• The paper demonstrates that increasing humanlike design enhances anthropomorphic perceptions via both surface cues and conversational sociability.
• Using cross-national studies and a factorial experiment, the research shows that heightened anthropomorphism does not uniformly boost user engagement.
• The behavioral trust game reveals that in some cultures, machine-like cues yield higher trust than overtly humanlike features.

Humanlike AI Design and Its Global Impact on Anthropomorphism, Engagement, and Trust

Introduction

This paper investigates how explicit design choices that render AI systems more humanlike affect users' perceptions, engagement, and trust across different cultural contexts. By manipulating two orthogonal axes—Design Characteristics (DC: surface-level humanlike cues and behavioral idiosyncrasies) and Conversational Sociability (CS: the system's propensity for empathetic, emotionally warm, or "human" interaction)—the authors systematically quantify the multi-faceted psychological and behavioral impact of anthropomorphic AI features.

Methodology

Two large-scale cross-national studies serve as the empirical foundation. In Study 1, qualitative coding and taxonomy-based analysis are applied to free text feedback after user-AI interactions involving food conversations, distributed across ten countries and multiple languages. Study 2 introduces a four-cell factorial experimental framework manipulating DC and CS, and evaluates effects via direct user ratings and a behavioral trust game. Participants represent diverse nations with varying cultural, religious, and linguistic backgrounds, and analyses account for demographic covariates and prior AI experience.

The robust construction of treatment prompts in Study 2 isolates DC and CS effects effectively: DC manipulation involves name usage, colloquial language, and conversational variability; CS manipulation modulates emotional warmth, empathy, and self-reflection. The trust game is designed with explicit monetary incentives, measuring actual behavioral trust rather than self-reported attitudes.

Key Findings and Numerical Results

Anthropomorphism

Firm evidence demonstrates that increased humanlike DC—distinctive names, pronouns, colloquial turns, emotive language—increases explicit user attribution of anthropomorphic qualities. CS further amplifies perceived social-emotional resonance. However, this anthropomorphic perception is not universal: substantial cross-country and within-country heterogeneity emerges, tightly correlated with cultural $F_{st}$ distance from the USA and other demographic factors.

Engagement and Trust

Contrary to intuitive design theory, maximizing anthropomorphism via DC and CS does not uniformly increase user engagement metrics or trust scores. Engagement, operationalized as conversational length, depth, and rating scales, varies idiosyncratically by treatment and cultural group. Higher anthropomorphism sometimes reduces engagement, especially in cultures distant from prototypical Western norms.

The incentivized trust game yields salient, contradictory findings: treatments producing the highest explicit anthropomorphism do not consistently maximize trust-based monetary transfers, and in some groups, emotionally neutral or "machine-like" designs lead to greater behavioral trust. This challenges dominant hypotheses in the HCI and social robotics domains linking anthropomorphism directly to trust.

Inter-rater Reliability and Automated Labeling

Human agreement (Cohen's Kappa) on anthropomorphic theme coding is highest on easily recognizable traits (friendliness, warmth, response speed), but drops sharply on subtler or culturally modulated traits such as intelligence and agency. The Gemini LLM used for automated labeling achieves a high $F_1$ score on major labels but struggles with nuanced categories, suggesting boundary cases remain problematic for ML-based sociocultural evaluation at scale.

Theoretical and Practical Implications

The results foreground that anthropomorphism is context-sensitive, not a monolithic psychological phenomenon. AI system designers should consider that cues designed to support anthropomorphic attributions may not reliably produce desired outcomes in engagement or trust and may have negative or paradoxical effects across cultures or user groups. The data also question the dominant Western-centric design logic in conversational AI and highlight the need for localized or personalized approaches.

For practitioners, the intricate mapping between DC, CS, and target behavioral metrics (engagement, trust, compliance) suggests that optimizing for one may not produce uniform improvements in others and may yield unintended adverse effects. The trust game results underscore the need to empirically test behavioral correlates of trust rather than rely on self-report or superficial engagement metrics.

Future Directions

The global divergence in outcomes calls for further investigation into the mechanisms by which anthropomorphic cues interact with user values, prior expectations, and local social norms. Expanding the taxonomy beyond predefined labels, improving ML models for culturally adaptive feedback analysis, and developing flexible, user-aware prompt engineering strategies will be critical. Longitudinal studies examining sustained engagement and trust evolution after repeated human-AI contact are warranted. There are also implications for regulatory policy regarding transparency and disclosure of AI "personality" traits in critical applications such as finance, healthcare, or governmental services.

Conclusion

This paper establishes that humanlike AI design robustly increases anthropomorphic perception but does not yield universally positive or consistent effects on engagement or trust, with substantial cross-cultural divergence. Theoretical models linking anthropomorphism and trust must be re-examined in light of these results, and practical AI design should favor empirically grounded, context-specific strategies rather than monolithic anthropomorphic enhancement (2512.17898).