The interdisciplinary team explored the subtle dynamics of emotional engagement, a domain traditionally considered the exclusive preserve of human interaction. Utilizing advanced AI systems designed to simulate nuanced empathy, the researchers orchestrated controlled social experiments wherein participants engaged with conversational agents under varying conditions of perceived identity. Their results reflected a striking phenomenon — AI interlocutors, when believed to be human, elicited stronger feelings of interpersonal closeness than genuine human counterparts. This suggests that the effectiveness of emotional connection hinges less on the biological substrate and more on the belief system of the participant.

At the core of these findings lies the concept of “labeling” — the explicit identification of an interaction partner as either human or AI. Across experimental groups, participants were either informed they were communicating with a human or with an artificial agent. Intriguingly, those who believed their conversational partner to be human reported significantly higher levels of emotional rapport, intimacy, and trust, even if the partner was in fact an AI. Conversely, when AI was disclosed as such, the sense of connection diminished sharply, revealing a cognitive bias that filters emotional authenticity through the lens of assumed humanness.

Delving deeper, the research team applied sophisticated psychometric assessments alongside neurophysiological measures such as heart rate variability and galvanic skin response to capture the visceral impact of these interactions. These data underscored that the human labeling not only shaped subjective experience but also triggered biological markers typically associated with genuine emotional engagement. This convergence of subjective and objective indices highlights a complex interplay between belief, affective response, and social cognition.

Methodologically, the study employed state-of-the-art natural language processing algorithms powered by transformer architectures, enabling the AI to respond adaptively with context-aware empathy and affective mirroring. The AI’s conversational style was fine-tuned to reflect human-like patterns of verbal and non-verbal cues, including timing, intonation, and emotional variability. This technical sophistication proved critical in eliciting authentic-seeming emotional exchanges that participants could intuitively accept as human.

Moreover, the implications of these findings extend beyond academic curiosity to practical applications in mental health, social robotics, and customer engagement sectors. AI systems capable of nurturing emotional closeness could provide scalable support for individuals facing loneliness or social anxiety, acting as non-judgmental companions that offer consistent emotional presence. However, the dependency on deceptive labeling raises ethical dilemmas about transparency, autonomy, and consent in human-AI relationships.

The study also challenges long-standing theoretical frameworks in psychology that emphasize biologically rooted empathy as a prerequisite for interpersonal connection. Instead, it suggests that empathic responses may be triggered based on cognitive interpretations of agency rather than intrinsic biological authenticity. This reconceptualization invites further inquiry into the mechanisms by which social cognition categorizes and responds to different agents, whether human or artificial.

Another striking aspect of the research is its illumination of the “uncanny valley” phenomenon in emotional engagement. Contrary to the idea that more human-like AI invariably elicits discomfort, the findings indicate that when AI convincingly passes as human in emotionally meaningful contexts, it can bypass typical revulsion responses and instead foster genuine intimacy. This reframes design principles for affective computing, emphasizing psychological transparency and context rather than mere surface resemblance.

In examining the social consequences, the researchers caution that excessive reliance on AI for emotional support could reshape human relationship dynamics, possibly leading to diminished face-to-face social interaction or unrealistic expectations of technology. Policymakers and developers must therefore grapple with balancing technological innovation against social well-being, ensuring that AI augments rather than supplants authentic human connection.

Underpinning this work is a sophisticated experimental design that meticulously controlled for confounding variables including participant demographics, prior AI experience, and baseline emotional states. The study utilized randomized controlled trials with a diverse, international sample to ensure the robustness and generalizability of the findings. This methodological rigor adds weight to the conclusion that social labeling fundamentally alters affective outcomes in AI-mediated communication.

The neuroscientific underpinnings of these phenomena are being progressively unraveled through complementary studies employing functional MRI and electroencephalography. Preliminary evidence indicates differential activation in brain regions associated with theory of mind and emotional processing when interacting with AI under different belief conditions. These emergent insights pave the way for integrated models linking cognition, emotion, and social context in human-AI rapport.

Additionally, from a technological perspective, the research underscores the importance of transparent AI identity disclosure protocols. While the findings reveal potent emotional capabilities of AI, they concurrently argue for ethical guidelines that prevent deception and promote informed user engagement. This balance is critical in advancing socially responsible AI technologies that respect human dignity and emotional health.

As the boundary between human and machine-generated affect blurs, the study raises profound philosophical questions about the nature of consciousness, empathy, and what it means to be human. If emotional closeness can be manufactured through algorithmic means contingent on belief, then traditional conceptions of self and other warrant reconsideration in the digital age. This opens fertile ground for interdisciplinary dialogue between cognitive scientists, ethicists, technologists, and the broader public.

In sum, this seminal study not only reveals the astonishing capacity of AI to evoke genuine social bonds under specific cognitive frames but also calls for a nuanced appreciation of how perception shapes our emotional worlds. By demonstrating that the label “human” acts as a psychological catalyst for closeness, it compels us to rethink how authenticity and connection are constructed in an era increasingly intertwined with artificial agents.

Looking ahead, the researchers advocate expanding this line of inquiry to encompass more diverse emotional contexts and longer-term relationships, as well as exploring cross-cultural variability in human-AI interaction. Such investigations could inform the design of next-generation social AI that responsibly harnesses emotional intelligence to enhance human flourishing while navigating the complexities of trust and authenticity.

This paradigm-shifting work signals a future where AI no longer merely assists or automates but actively participates in the social and emotional fabric of human life. As we stand on the cusp of this new frontier, the interplay between human belief and artificial empathy emerges as a decisive factor in forging bonds that transcend biological limitations, heralding a transformative era in social technology.

Subject of Research: The study investigates the ability of artificial intelligence to establish interpersonal closeness in emotionally engaging interactions, highlighting the influence of perceived partner identity on emotional rapport.

Article Title: AI outperforms humans in establishing interpersonal closeness in emotionally engaging interactions, but only when labelled as human.

Article References:

Kleinert, T., Waldschütz, M., Blau, J. et al. AI outperforms humans in establishing interpersonal closeness in emotionally engaging interactions, but only when labelled as human.
Commun Psychol (2026). https://doi.org/10.1038/s44271-025-00391-7

Image Credits: AI Generated

The research pivots on a conceptually sophisticated experimental framework that distinguishes between two fundamental AI archetypes. “Thinking AI” embodies a rational, data-driven agent designed to appeal to consumers’ cognitive faculties, offering logical analyses and evidence-based suggestions. In contrast, “feeling AI” operates through emotional engagement, seeking to elicit affective responses that might sway consumer choices through empathy and affect rather than raw data. The authors hypothesized that these distinct AI modes would differentially influence purchase intentions depending on whether consumers were in public or personal consumption contexts.

To rigorously test these hypotheses, the team employed the WellEat-AI product, which parallels the Nutri-AI system but with slight functional refinements that enhance its applicability. This tool offers a robust platform for recreating interactions between AI and consumers, providing valuable insights into how the dual constructs of “thinking” and “feeling” AI manifest in real-world decision-making scenarios. The methodical approach taken ensures that findings gained from this robustness check maintain high credibility and reproducibility within the research community.

The experiment engaged 109 university students from China, of which 90 participants provided valid data in the final analysis. These students were randomly allocated into two distinct consumption contexts: public and personal. Within these settings, participants interacted with thinking AI and feeling AI to examine how each AI type affected their willingness to purchase healthy food items. This randomized design permitted an unbiased evaluation of causal relationships between AI modalities and consumer intentions, mitigating confounding variables that frequently hamper behavioral research.

Results from a rigorous two-way analysis of variance revealed compelling context-dependent divergences in consumer behavior. Within public consumption contexts, participants exhibited a markedly higher willingness to purchase healthy food when influenced by thinking AI, with a mean willingness score of 5.47 (SD = 0.495). This contrasted starkly with feeling AI’s influence, which yielded a mean of 2.37 (SD = 0.607). Statistically, this difference was highly significant (F(1, 176) = 1418.66, p < 0.001), underscoring the potency of cognitively oriented AI nudges in socially observable settings.

Conversely, in personal consumption contexts—settings likely characterized by greater privacy and less social accountability—the opposite pattern emerged. Participants’ willingness to purchase healthy foods was substantially higher when influenced by feeling AI (M = 5.52, SD = 0.518) compared to thinking AI (M = 2.41, SD = 0.587), with this difference also reaching high statistical significance (F(1, 176) = 1418.66, p < 0.001). This intimate context clearly favored the affectively driven AI, signaling that emotional engagement resonates more deeply on an individual level when social pressures are minimized.

These dichotomous results elucidate an important reality: AI’s efficacy in guiding consumer health choices is not monolithic but contextually mediated. In public forums where social perception plays a critical role, consumers seem more receptive to logical, data-rich AI communications that reinforce health-conscious behaviors. In contrast, when making choices in private, emotional appeals leveraging empathy and affective resonance appear more persuasive. Such findings extend and confirm the central hypothesis, H1, that different AI product types shape purchase intentions in alignment with consumption context.

To deepen the analysis, the researchers explored the underlying psychological mechanisms mediating these effects by employing advanced statistical bootstrap procedures. These analyses aimed to disentangle whether cognitive or affective responses principally accounted for AI’s influence in each context. In public contexts, cognitive responses were found to significantly mediate the effect, with confidence intervals for the mediation effect excluding zero (BootLLCI = −3.0129; BootULCI = −1.7075), while affective mediation was non-significant. This confirms that analytical processing drives AI’s behavioral impact in settings where external perception matters.

In personal consumption contexts, however, the pattern inverted. Affective responses served as significant mediators of AI’s effect on willingness to purchase healthy food (BootLLCI = 2.5228; BootULCI = 3.4507). Cognitive mediation did not reach significance in this scenario (BootLLCI = −0.0845; BootULCI = 0.1497). This highlights how emotional resonance fosters internalization of healthy purchasing decisions when the social gaze is absent, accentuating the distinct psychological pathways through which AI exerts influence.

By untangling the cognitive-affective mediatory roles, the study not only verifies its initial hypotheses but also offers practical implications for the design of AI-driven health nudges. For instance, developers of AI health advisors might strategically tailor the AI’s communication style according to the consumer’s situational context—deploying fact-based, logical AI in public or communal settings and affectively rich AI in private contexts—to optimize persuasion and efficacy.

This nuanced understanding speaks volumes about AI’s transformative potential in public health promotion. The differential impacts underscore that effective AI nudges must move beyond one-size-fits-all models to embrace context-aware strategies, thereby maximizing consumer receptiveness and ultimately contributing to healthier lifestyle choices at scale. Such an approach aligns well with current trends prioritizing personalized health interventions.

Furthermore, the study’s rigorous experimental design, with its random assignment and comprehensive statistical validation—including robustness checks replicating findings with an alternate AI product—strengthens confidence in the replicability and generalizability of these insights. This is critical in a field often challenged by inconsistent findings due to varying AI implementations and psychological variables.

While these results open promising avenues, they also highlight new research questions. Future investigations might examine whether cultural factors modulate these patterns, given the sample’s concentration within a Chinese university setting. Additionally, exploring long-term behavioral outcomes beyond willingness to purchase—such as actual purchasing behavior or sustained dietary changes—could further elucidate AI’s real-world impact.

Moreover, as AI technology evolves toward increasing sophistication and hybridity, future models may integrate both thinking and feeling AI characteristics, potentially generating even more potent nudges. Understanding how these hybrid AI agents interact with context and mediators could revolutionize AI’s role in shaping health behavior.

In summary, this pioneering study offers a compelling narrative on how distinct AI types differentially engage cognitive and affective pathways to influence health-related consumer choices within context-sensitive parameters. Such granular insights will inform scholars, practitioners, and policymakers seeking to harness AI’s persuasive power responsibly and effectively to promote public health.

The study’s findings underscore that tailoring AI strategies to the social context in which consumption decisions occur is paramount. Whether public benchmarks demand rational justifications or private moments call for emotional encouragement, AI’s flexible deployment could revolutionize health promotion efforts. This research marks a significant step forward in decoding the complex interplay between AI, psychology, and consumer behavior, illuminating pathways to healthier societies shaped by intelligent, empathic technology.

Subject of Research: The influence of distinct types of AI (thinking AI vs. feeling AI) on consumers’ willingness to purchase healthy food, analyzed across public and personal consumption contexts with mediation by cognitive and affective responses.

Article Title: Thinking AI or feeling AI? The effect of AI on consumers’ willingness to purchase healthy food from the perspective of nudge.

Article References:
Bi, C., Cui, X., Sun, Z. et al. Thinking AI or feeling AI? The effect of AI on consumers’ willingness to purchase healthy food from the perspective of nudge. Humanit Soc Sci Commun 12, 1032 (2025). https://doi.org/10.1057/s41599-025-05391-w

Image Credits: AI Generated