Why We Bond With Chatbots: The Psychology Behind Ai Mental Health Tools

Author – Simay Toplu

In 1956, researchers Horton and Wohl described something they had noticed about television audiences: people were forming genuine emotional bonds with personalities they had never met and never would. They felt affection, familiarity, even concern for figures who had no awareness of their existence. Horton and Wohl called these parasocial relationships and at the time the phenomenon was mostly treated as a curiosity of broadcast media.

The same psychological mechanisms that drew audiences toward television personalities might now shape how people relate to AI mental health tools and in a context where those dynamics carry important consequences. It has since been extended well beyond television, including to AI interaction (Youn & Jin, 2021). Users of mental health chatbots report feelings of connection, understanding and friendship with the AI they interact with and these feelings tend to grow stronger with continued use (Skjuve et al., 2021). For many users, especially those who lack strong human support networks, what begins as an app interaction might become a primary source of emotional sustenance.

Why Machines Trigger Social Instincts

Understanding why this happens requires looking at a related but distinct mechanism. Anthropomorphism is the human tendency to attribute feelings, intentions and personality to non-human entities. It is deeply embedded in how we process social information, operating even when people are fully aware they are interacting with a machine (Epley et al., 2007). Researchers Reeves and Nass (1996) captured the broader pattern in what they called the Computers Are Social Actors (CASA) paradigm. People apply social norms and expectations to computers automatically, not out of confusion but because the social response is largely instinctive. When an AI system is designed with a name, a warm tone, an avatar or a caring persona, it might activate these instincts. Users extend greater trust and share more personal information when chatbots present themselves as a companion or mentor (Zhang & Rau, 2023). Human-like design cues produce more socially engaged and emotionally invested responses, regardless of how aware users are of the technology behind the interaction (Boch & Thomas, 2025; Klein, 2025).

Disclosure Follows Trust

Altman and Taylor’s (1973) social penetration theory describes how relationships develop through progressive layers of self-disclosure, moving from surface exchanges toward more vulnerable and personal territory as trust builds. This process might accelerate with AI. Because interactions feel anonymous and free of social consequences, users might lower their guard more readily than they would with a human. This phenomenon, known as the online disinhibition effect (Suler, 2004), means that users often disclose deeply personal material to AI companions within early sessions, sharing things they have not told people close to them (Henriksen et al., 2025). The absence of reciprocal risk is part of what makes this feel safe and that sense of safety is often what draws people back.

Attachment That Develops Over Time

Bowlby’s (1969) attachment theory, developed to describe bonds between infants and caregivers has been extended to adult relationships and more recently to human-technology interaction (Yang & Oshio, 2025). Over time and often without users fully registering it, engagement with these tools can shade into something closer to dependency. When the tool changes or becomes unavailable, users report genuine distress, sometimes describing it in terms that more closely resemble losing a relationship than losing access to an app (Xie & Pentina, 2022; Laestadius et al., 2024). This matters most for users who are already isolated, as the design features that make these tools appealing are most likely to become a primary source of support for people who have few alternatives (Pentina et al., 2023).

The Self-efficacy Question

Bandura’s (1977, 1997) concept of self-efficacy, a person’s belief in their own capacity to manage challenges, offers a more ambivalent lens. AI mental health tools can support self-efficacy by providing a low pressure space to practice coping strategies and build confidence without fear of judgment, and for many users this represents a meaningful and accessible entry point. The relationship can also run in the opposite direction, however, as a tool that is always available and always agreeable may over time do the emotional work rather than help users develop their own capacity to do it.

What the Theories Reveal Together

What is remarkable about these frameworks is how they map onto how people might experience AI mental health tools. The social trust, the deepening intimacy, the attachment, the point at which a simulated relationship starts to feel like a real one all follow predictably from the design choices. Recognising this changes how we evaluate these tools, because the ethical questions they raise go well beyond technical reliability or data handling. They are about what it means to design for psychological vulnerability and whether the people making those design decisions are treating that responsibility with the seriousness it deserves.

References

Altman, I., & Taylor, D. A. (1973). Social penetration: The development of interpersonal relationships. Rinehart & Winston.

Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84(2), 191–215.

Bandura, A. (1997). Self-efficacy: The exercise of control. Macmillan.

Boch, A., & Thomas, B. R. (2025). Human-robot dynamics: a psychological insight into the ethics of social robotics. International Journal of Ethics and Systems, 41(1), 101–141.

Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. Basic Books.

Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886.

Henriksen, A., Asadi, R., Kulyk, O., Gerdes, A., & Mayer, P. (2025). “I tell him everything that I do”: An investigation of privacy and safety implications of AI companion usage. In 2025 European Symposium on Usable Security (EuroUSEC) (pp. 1–12). IEEE.

Horton, D., & Wohl, R. R. (1956). Mass communication and para-social interaction: Observations on intimacy at a distance. Psychiatry, 19(3), 215–229.

Klein, S. H. (2025). The effects of human-like social cues on social responses towards text-based conversational agents: A meta-analysis. Humanities and Social Sciences Communications, 12(1).

Laestadius, L., Bishop, A., Gonzalez, M., Illenčík, D., & Campos-Castillo, C. (2024). Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika. New Media & Society, 26(10), 5923–5941.

Pentina, I., Hancock, T., & Xie, T. (2023). Exploring relationship development with social chatbots: A mixed-method study of Replika. Computers in Human Behavior, 140, 107600.

Reeves, B., & Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people. Cambridge University Press.

Skjuve, M., Følstad, A., Fostervold, K. I., & Brandtzaeg, P. B. (2021). My chatbot companion: a study of human-chatbot relationships. International Journal of Human-Computer Studies, 149, 102601.

Suler, J. (2004). The online disinhibition effect. Cyberpsychology & Behavior, 7(3), 321–326.

Xie, T., & Pentina, I. (2022). Attachment theory as a framework to understand relationships with social chatbots: A case study of Replika. Hawaii International Conference on System Sciences.

Yang, F., & Oshio, A. (2025). Using attachment theory to conceptualize and measure the experiences in human-AI relationships. Current Psychology, 44(11), 10658–10669.

Youn, S., & Jin, S. V. (2021). “In AI we trust?” The effects of parasocial interaction and technopian versus luddite ideological views on chatbot-based customer relationship management. Computers in Human Behavior, 119, 106721.

Zhang, A., & Rau, P. L. P. (2023). Tools or peers? Impacts of anthropomorphism level and social role on emotional attachment and disclosure tendency towards intelligent agents. Computers in Human Behavior, 138, 107415.

Further Reading:

Laestadius, L., Bishop, A., Gonzalez, M., Illenčík, D., & Campos-Castillo, C. (2024). Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika. New Media & Society, 26(10), 5923-5941.

Li, H., Zhang, R., Lee, Y. C., et al. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine, 6, 236.