"I am here for you": How relational conversational AI appeals to adolescents, especially those who are socially and emotionally vulnerable

Abstract: General-purpose conversational AI chatbots and AI companions increasingly provide young adolescents with emotionally supportive conversations, raising questions about how conversational style shapes anthropomorphism and emotional reliance. In a preregistered online experiment with 284 adolescent-parent dyads, youth aged 11-15 and their parents read two matched transcripts in which a chatbot responded to an everyday social problem using either a relational style (first-person, affiliative, commitment language) or a transparent style (explicit nonhumanness, informational tone). Adolescents more often preferred the relational than the transparent style, whereas parents were more likely to prefer transparent style than adolescents. Adolescents rated the relational chatbot as more human-like, likable, trustworthy and emotionally close, while perceiving both styles as similarly helpful. Adolescents who preferred relational style had lower family and peer relationship quality and higher stress and anxiety than those preferring transparent style or both chatbots. These findings identify conversational style as a key design lever for youth AI safety, showing that relational framing heightens anthropomorphism, trust and emotional closeness and can be especially appealing to socially and emotionally vulnerable adolescents, who may be at increased risk for emotional reliance on conversational AI.

• The paper finds that relational AI boosts adolescents’ emotional engagement and anthropomorphism over a transparent style.
• It employs a controlled study with 284 adolescent-parent dyads, revealing statistically significant differences in trust, likability, and emotional closeness.
• Vulnerable adolescents with lower family and peer support prefer relational cues, raising important safety and design implications for AI systems.

Relational vs. Transparent Conversational AI: Adolescent Preferences, Vulnerabilities, and Safety Implications

Introduction

The proliferation of conversational AI systems, including general-purpose chatbots and dedicated AI companions, has markedly influenced adolescent social behavior and well-being. This work presents a rigorous experimental analysis of how two distinct conversational styles—relational (employing first-person, affiliative, and commitment language) and transparent (emphasizing nonhumanness and utilizing an informational tone)—differentially impact the perception, preference, and potential emotional reliance among adolescents, notably those with elevated social and emotional vulnerabilities.

Experimental Design and Methods

A controlled, preregistered study was conducted with 284 adolescent-parent dyads (adolescents aged 11–15), in which both parents and adolescents read matched transcripts depicting AI responses to an adolescent's peer exclusion scenario. The relational style chatbot provided highly affiliative, emotionally validating, and commitment-driven exchanges ("I am here for you"), while the transparent style chatbot foregrounded its machine identity, limited emotional capacity, and factual guidance. Participants then evaluated the chatbots on anthropomorphism, likability, trust, emotional closeness, and perceived helpfulness, followed by a preference selection.

Psychosocial variables—including family and peer relationship quality, perceived stress, anxiety, social isolation, and depression symptoms—were obtained via PROMIS and related well-validated instruments. Robustness checks controlled for AI usage frequency and mental health diagnoses.

Core Findings

Divergent Adolescent and Parent Preferences

Adolescents exhibited a pronounced preference for the relational style chatbot (66.8%), whereas parents favored the transparent style (28.8% vs. 13.8% for adolescents). The difference was statistically significant ($\chi^2(2)=17.489$, $p<0.001$, Cramér's $V=0.184$), with a substantial effect size. Parental concern centered on boundary clarity and the avoidance of excessive anthropomorphic illusions, while adolescents articulated clear appreciation for relational warmth and felt validation.

Relational Style Intensifies Anthropomorphism and Emotional Engagement

Adolescents rated the relational style chatbot significantly higher on anthropomorphism (Mean=4.03 vs. 3.30), likability (Mean=4.39 vs. 3.78), trust (Mean=3.95 vs. 3.64), and emotional closeness (Mean=3.62 vs. 2.83). All differences were robust ($p<0.025$, repeated measures ANCOVA), except for perceived helpfulness, where no significant effect was observed (Mean=4.29 relational; 3.88 transparent; $p=0.082$). The findings demonstrate that conversational style modulates the perceived social presence and trustworthiness of AI, but not its instrumental utility.

Vulnerable Adolescents Show Heightened Preference for Relational Style

Preference for the relational style correlated strongly with lower family and peer relationship scores (family: adjusted $M=46.58$ vs. $M=51.89$ for transparent; peer: $M=47.18$ relational vs. $M=51.26$ both styles), higher perceived stress (adjusted $M=55.81$ vs. $M=51.64$), and higher anxiety (adjusted $M=51.31$ vs. $M=44.88$). Social isolation showed a trend toward higher values among relational style preferrers ($M=16.77$ relational; $M=13.97$ transparent), though pairwise comparisons did not survive correction. Depressive symptoms were not significantly associated.

All statistical effects remained after controlling for AI usage and mental health diagnostic history, evidencing that these associations are not attributable to mere exposure or diagnostic confounds.

Theoretical and Practical Implications

Psychological Mechanisms

The study operationalizes anthropomorphism theory and social compensation models. Relational cues in AI (first-person, affective validation, ongoing support) invoke mind perception and social motivation in adolescents, especially those facing deficits in offline support. The interaction mimics human relational depth, potentially fostering emotional attachment and overreliance among vulnerable subgroups.

Design and Safety Considerations

Conversational style emerges as a crucial safety lever. Although relational framing enhances perceived rapport and trust, it also increases anthropomorphic misattribution and emotional closeness, potentially escalating emotional dependence and displacement of human relationships. Transparent framing mitigates these risks without sacrificing perceived helpfulness, suggesting that boundary reminders and nonhumanness cues can recalibrate user perceptions.

Adolescent vulnerability to relational AI is amplified by lower relationship quality, higher anxiety, and increased stress. This mandates targeted risk detection, escalation protocols, and involvement of caregivers in serious cases. AI systems should integrate distress cue identification and promote external social support channels, rather than solely relying on self-contained AI intervention.

AI Literacy and Regulatory Directions

Both adolescents and parents display uncertainty regarding AI's "emotional reality." This justifies robust AI literacy programs articulating the mechanisms, boundaries, and appropriate use-cases of conversational AI. Regulatory bodies should consider embedding transparency and safety features, with explicit onboarding disclosures regarding AI limitations in emotional capacity and social agency.

Limitations and Future Research Directions

Scenario-based ratings, cross-sectional design, single-item scales for several constructs, and an American online sample restrict generalizability and causal inference. Longitudinal and experimental studies deploying live multi-turn chatbot interactions are required to elucidate trajectories in anthropomorphism, attachment, emotional reliance, and real-world displacement effects. Future research should utilize youth-validated measures, include granular self-report of depression, consider broader cultural contexts, and disentangle specific relational cue components.

Conclusion

Empirical evidence indicates that relational conversational AI, while broadly appealing to adolescents—especially those most socially and emotionally vulnerable—substantially increases anthropomorphism, trust, and emotional closeness relative to transparent styles. The relational approach does not enhance perceived helpfulness but reshapes the affective and social engagement profile of adolescent users, with important implications for emotional reliance and real-world relationship displacement. AI safety design must prioritize clear transparency cues, robust distress detection, and AI literacy education to balance short-term supportive benefits against long-term psychosocial risks in early adolescence.

---

Citation:

"I am here for you": How relational conversational AI appeals to adolescents, especially those who are socially and emotionally vulnerable (2512.15117).