In an era when millions of young people turn to conversational artificial intelligence for comfort during moments of loneliness, sadness, and anxiety, a new study reveals that two deceptively simple design choices—how formally a chatbot speaks and whether it is labeled as “generative AI” or a “traditional chatbot”—can meaningfully shape how emotionally supported users feel. The research, published in Information Systems Frontiers by Jian-Ren Hou and Maria Bellaniar Ismiati of National Cheng Kung University in Taiwan, offers some of the most granular evidence to date on the psychological mechanics of human-chatbot interaction among one of the most digitally connected and emotionally vulnerable student populations: international students studying far from home.
The research team built a generative AI chatbot named Airbud and presented it to 450 international students under different identity labels—some participants were told they were interacting with a generative AI system, while others were led to believe they were using a conventional, rule-based chatbot. Within those experimental conditions, the chatbot was further configured to communicate in either a formal or an informal style, ranging from polished, structured, and professionally worded responses to casual, colloquial exchanges. This dual manipulation allowed the researchers to isolate the independent and combined effects of machine identity and linguistic register on users’ perceptions of emotional support.
The methodological backbone of the study is structural equation modeling, a statistical framework that allows researchers to test networks of hypothesized relationships among latent psychological constructs—variables like “perceived warmth,” “social presence,” and “perceived tie strength” that cannot be observed directly but are inferred from patterns in survey responses. Using this approach, the team discovered that a formal communication style actually enhances perceived tie strength, the sense of closeness and connection a user feels toward the chatbot. This finding runs counter to a popular intuition in conversational design that casual, friendly language is always warmer and more bonding. For international students—many of whom navigate daily life in a second or third language—a formal register may signal competence, reliability, and respect, qualities that translate into feeling genuinely attended to rather than patronized.
Perceived warmth emerged as a critical amplifier in the study’s models. When users perceived the chatbot as warm, the positive effects of the communication style on social presence—the feeling that a real social partner is “there” on the other side of the screen—and on perceived tie strength were significantly strengthened. Warmth also amplified emotional relief, the sense that a burden had been lifted through the interaction. In psychological terms, warmth appears to act as a moderator that converts competent, formal communication into something that feels personally caring. A chatbot can be grammatically flawless and structurally formal, but unless users detect warmth in its tone, that formality risks reading as cold corporate scripting rather than genuine attentiveness.
The study also examined how interaction satisfaction—the user’s overall evaluation of the conversational experience—shapes what the researchers call perceived chatbot support: the degree to which users believe the chatbot helps with loneliness, sadness, and anxiety. Importantly, the authors are careful to define this construct as perceived support efficacy rather than actual clinical improvement or measurable emotional change. In other words, the study captures how supported students feel after talking to the machine, not whether the interaction produced durable therapeutic outcomes. This distinction matters for interpreting the findings responsibly: a chatbot may reliably generate a subjective sense of relief without functioning as a substitute for professional mental health care, and the researchers’ framing acknowledges that boundary explicitly.
Perhaps the most intriguing result comes from the multi-group analysis comparing the two identity labels. The researchers found that formal communication style is less strongly related to perceived judgment when the chatbot carries a generative AI label. Perceived judgment refers to the user’s fear of being negatively evaluated—a construct closely related to the fear of negative evaluation that has long been studied in social anxiety research. For many users, disclosing emotional struggles to another entity carries an implicit social risk: the worry of being criticized, pitied, or misunderstood. The findings suggest that when users know they are talking to a generative AI system, the psychological stakes of that judgment appear to drop, making even formal language feel less threatening. A generative AI label may act as a kind of psychological buffer, signaling that the interlocutor is a machine without social agendas, gossip networks, or opinions of its own.
This interpretation aligns with a foundational body of research in human-computer interaction. The classic “computers are social actors” paradigm, established by Clifford Nass and colleagues in the 1990s, demonstrated that people reflexively apply social rules to machines—politeness norms, reciprocity, and yes, fear of judgment. But the new study suggests that the generative AI label may partially modulate that reflex, selectively weakening the social-evaluative dimension while preserving the relational benefits of the interaction. It is a nuanced picture: users do not simply treat a chatbot as either fully social or fully mechanical; instead, the label calibrates which social responses are activated.
The choice of international students as the study population is not incidental. This group faces a distinctive constellation of stressors—cultural dislocation, language barriers, academic pressure, separation from family support networks, and often limited access to culturally attuned mental health services. Prior research has documented elevated loneliness and reluctance to seek professional psychological help among international students, driven in part by stigma and unfamiliarity with host-country counseling systems. Chatbots occupy an appealing niche in this landscape: they are available around the clock, incur no appointment wait, speak without fatigue, and carry no social consequences. If a student is worried that confessing homesickness to a human counselor might feel embarrassing, confiding the same feeling to a well-designed chatbot sidesteps much of that friction.
The technical measurement strategy behind the study reinforces the credibility of its conclusions. The researchers drew on validated psychometric instruments for constructs such as loneliness, tie strength, and social presence, and evaluated their structural model using partial least squares-based structural equation modeling techniques, a family of methods well established for exploratory research on perceptual constructs. Multi-group analysis, the technique used to compare the generative AI and traditional chatbot conditions, tests whether the estimated path coefficients differ significantly between subgroups—in this case revealing that the relationship between formality and perceived judgment is genuinely moderated by the identity label rather than being an artifact of sampling noise.
For the designers of conversational systems, the practical implications are immediate. First, communication style should be treated as a deliberate design variable, not an afterthought. Formality can build trust and tie strength, particularly when users value competence and respect. Second, warmth cues—empathetic phrasing, acknowledgment of the user’s feelings, attentive follow-ups—are not decorative; they are mechanistically important, strengthening the pathway from style to presence and relief. Third, transparency about a system’s generative AI nature, increasingly required by regulation and platform policy, may carry an unexpected psychological dividend: rather than undermining emotional connection, the label may reduce perceived judgment and make users more comfortable disclosing difficult feelings.
At the same time, the study’s authors and the broader literature counsel caution. The growing ecosystem of AI-driven mental health tools—from large-scale commercial therapy chatbots to campus wellness assistants—rests heavily on users’ subjective sense of being helped, which this study shows can be systematically shaped by superficial design cues. A label or a register of speech that boosts perceived support does not necessarily produce clinical benefit, and researchers have documented both algorithm aversion, in which people abandon automated systems after observing errors, and the placebo-like effects of AI branding in human-computer interaction. Responsible deployment will require pairing the persuasive design levers identified here with rigorous outcome evaluation, clear escalation pathways to human professionals, and safeguards for users in genuine crisis.
What the Airbud experiments ultimately demonstrate is that the emotional texture of a chatbot conversation is a composite of engineering and psychology: the statistical signature of a language model’s word choices, the framing of its identity on the screen, and the perceptual machinery of the human mind interpreting both. For international students navigating the dislocation of study abroad, those small design decisions can determine whether a midnight conversation with a machine feels like hollow scripted dialogue or a genuine moment of relief. As universities and technology companies continue building AI companions for emotional well-being, this research provides a precise, evidence-based map of which levers matter—and a reminder that the difference between a chatbot that soothes and one that merely responds may lie in a label and a turn of phrase.
Subject of Research: How generative AI identity labels, formal communication style, and perceived warmth influence international students’ perceived emotional support in chatbot interactions.
Subject of Research: Technology and Engineering
Article Title: Generative AI Labeling, Communication Style, and Perceived Emotional Support in Chatbot Interactions among International Students
Article References: Hou, J.-R., & Ismiati, M. B. (2026). Generative AI Labeling, Communication Style, and Perceived Emotional Support in Chatbot Interactions among International Students. Information Systems Frontiers. https://doi.org/10.1007/s10796-026-10814-3
Image Credits: AI Generated
DOI: 10.1007/s10796-026-10814-3
Keywords: generative AI, chatbots, communication style, perceived warmth, emotional support, international students, social presence, tie strength, interaction satisfaction, chatbot identity label, perceived chatbot support, mental well-being
Cite Scienmag News
APA MLA Chicago
Glenn Wilkins. (September 4, 2026). AI Disclosure and Emotional Support in Chatbot Talks with International Students. Scienmag. https://scienmag.com/ai-disclosure-and-emotional-support-in-chatbot-talks-with-international-students/
Glenn Wilkins. “AI Disclosure and Emotional Support in Chatbot Talks with International Students.” Scienmag, 4 September 2026, https://scienmag.com/ai-disclosure-and-emotional-support-in-chatbot-talks-with-international-students/. Accessed 4 September 2026.
Glenn Wilkins. “AI Disclosure and Emotional Support in Chatbot Talks with International Students.” Scienmag. September 4, 2026. https://scienmag.com/ai-disclosure-and-emotional-support-in-chatbot-talks-with-international-students/
Copy citation Download RIS
Tags: AI chatbot designAI chatbot emotional supportAI disclosure and user trustAI-based loneliness and anxiety reliefchatbot design and user perceptionconversational AI and mental healthcross-cultural AI communicationcross-cultural AI interactionsdigital mental health support toolsemotional support for international studentsemotional well-being and AIformal vs informal chatbot communicationgenerative AI versus rule-based chatbothuman-chatbot emotional interactionhuman-chatbot interactionimpact of chatbot formality on user experienceinfluence of chatbot communication styleinternational students mental healthlabeling AI as generative or traditionalpsychological effects of AI disclosurepsychological impact of AI chatbotsvirtual emotional comfort for students abroadvirtual emotional support for international students
