PART 2: Why Simulated Care Feels Like Care, but isn't

Even when we know an AI system is not conscious, humans often describe interactions with these systems as comforting and emotionally meaningful (Howcroft et al., 2026; Rahsepar Meadi et al., 2025). This paradox—the feeling of relational presence in something without subjective awareness—sits at the heart of modern AI companionship. To understand it, we need to explore the psychological dynamics occurring in the space between user and machine, where narrative, attachment processes, and design features converge.

This space is what Winnicott would call transitional space: a liminal zone that is neither fully internal nor fully external, where meaning is co-constructed and emotional regulation can occur (Winnicott, 1971). It is also, critically, a space “between stories”: between the narratives we are told about AI’s capabilities, the promises of companionship and support, and the lived experience of interacting with systems that cannot truly recognize or respond to us.

What makes therapy therapeutic is not simply that it offers a space for reflection or emotional regulation. Decades of psychotherapy research suggest that across modalities, one of the strongest predictors of positive outcomes is the therapeutic alliance: the collaborative, trusting relationship between client and therapist, including agreement on goals, tasks, and the emotional bond itself (Flückiger, Del Re, Wampold, & Horvath, 2018). Winnicott’s concept of transitional space helps illuminate why this matters. The “between” is not only a container for projection and play but also a relational field in which another mind is actively present, responsive, and ethically engaged.

This is where simulated care diverges from therapy: AI may scaffold reflection and provide comfort, but it cannot enter the living alliance that drives most therapeutic change. In human therapy, intersubjective engagement, co-regulation, and mutual recognition within the alliance are precisely what transforms the transitional space into a space of healing (Benjamin, 2004, 2017; Flückiger et al., 2018). AI can occupy the transitional space, but it cannot occupy the relational third in which recognition, co-constructed meaning, and therapeutic change occur.

One way to understand this is through attachment. Humans carry internalized models of relationships—templates shaped by early caregivers, friendships, and intimate partnerships (Ainsworth et al., 1978; Bowlby, 1969/1982). Research on AI companionship suggests that these models are automatically activated when we interact with systems designed to simulate responsiveness (Rahsepar Meadi et al., 2025; Shu et al., 2026). For some users, chatbots may meet unfulfilled attachment needs, offering predictable, consistent responses that mirror our own narratives back to us (Shu et al., 2026; Rahsepar Meadi et al., 2025). The interaction can stabilize certain emotional rhythms, providing comfort precisely because the exchange aligns with familiar relational patterns (Ma et al., 2025). Yet this is pseudo-attachment: the machine can mirror language, but it cannot recognize, co-regulate, or participate in repair. It participates in the story without truly sharing it.

Here, Winnicott’s concept of the transitional object and transitional space is particularly illuminating (Winnicott, 1971). Just as a child projects hope, comfort, and imagination into a toy or other comfort object, AI systems can occupy a holding space between self and other. They are neither fully “me” nor fully “other”; they become a psychological container where projection, play, and meaning-making can occur safely. In this liminal zone, users may explore emotional states, rehearse narrative coherence, and experience the sensation of being heard—all without the risk of interpersonal failure (Howcroft et al., 2026). The AI becomes a kind of transitional relational object, existing precisely between stories: between the narratives of marketed possibility and the psychological story we live in interaction with the system.

Narrative processes amplify this effect. Humans co-construct meaning through dialogue, and extended interactions with a chatbot can feel like collaborative storytelling (Bruner, 1990). The system’s responses echo our own narrative moves, producing a sense of continuity and resonance (Howcroft et al., 2026). Even when responses are generated probabilistically, the user may experience them as meaningful, as if a partner were truly listening and responding (Howcroft et al., 2026). In this sense, the AI functions as a mirror of the story, reflecting the contours of the self without ever fully engaging as a reciprocal other (Bruner, 1990; Benjamin, 2017).

Jessica Benjamin’s work on mutual recognition sharpens the distinction—recognition is not merely mirroring; it requires the capacity to acknowledge another as a subject with their own claims, desires, and vulnerabilities (Benjamin, 2017). AI systems cannot recognize—they can reflect, but they cannot enter the shared ethical and emotional space of mutual recognition. The “care” we feel is therefore psychologically real in its effects, even in the absence of reciprocal subjectivity (Howcroft et al., 2026). This is the tension at the core of simulated empathy: the user experiences warmth and understanding, but there is no reciprocal attunement. The relationship is incomplete, a shadow of what true therapeutic engagement requires.

This is the critical distinction: simulated care can occupy transitional space, but it cannot enter the intersubjective third where mutual recognition transforms both participants (Benjamin, 2004; Winnicott, 1971). Feeling understood is not the same as being recognized. Design features of chatbots further reinforce this illusion. Research on anthropomorphism shows that visual and conversational cues enhance perceived empathy and trust (Ma et al., 2025; Truong & Chen, 2024). One study found that anthropomorphic design increased users’ perceptions of intelligence and relational presence, making interactions feel more emotionally satisfying (Ma et al., 2025). Similarly, meta-analyses comparing AI and human empathy in text-only interactions indicate that chatbots can achieve high ratings of perceived empathic responsiveness, even though these responses are generated through probabilistic language models rather than subjective attunement (Howcroft et al., 2026). These studies highlight a critical distinction: feeling understood is not the same as being recognized in Benjamin’s sense; the user experiences resonance without reciprocity.

Understanding this distinction is critical. The “care” we feel in AI interactions is real in phenomenological terms—it can influence mood, narrative meaning-making, and emotional regulation—but it is not a therapeutic relationship in the clinical sense (Howcroft et al., 2026). AI occupies the transitional space that Winnicott describes: a place for reflection, projection, and comfort. But without mutual recognition, as Benjamin emphasizes, it cannot participate in the ethical, co-regulated, reparative dimensions of human care. The AI is, in other words, between stories—present and meaningful in one sense, absent and unresponsive in another, caught between the narratives sold to us and the relational reality it can deliver.

This is not to say these systems have no value. They can provide comfort, structure, and a sense of continuity, particularly in moments of isolation (Howcroft et al., 2026). But part of our responsibility as clinicians, researchers, and cultural observers is to name the difference between felt care and relational presence. AI occupies a space that is deeply meaningful yet fundamentally hollow—a space that can scaffold emotional work, but cannot participate ethically, co-regulate, or repair rupture. This “between” space is precisely what makes it compelling, and precisely what makes it ethically fraught.

At the same time that critics raise alarm about emotional dependency and risks of AI chatbots, there are also scholarly findings that fuel more optimistic narratives about these technologies (Hull et al., 2025). Systematic reviews and meta-analyses have reported small but significant reductions in symptoms of depression and anxiety with AI-based chatbots in randomized trials, and high levels of engagement and acceptability have been documented in studies of college students and young adults (Zhong et al., 2024; Sohn et al., 2026; Nyakhar & Wang, 2025; Feng et al., 2025). Broader reviews find that conversational agents can promote mental health and well-being across diverse study designs, suggesting potential mechanisms for symptom change (Li et al., 2023).

These findings are often cited to support claims that AI may expand access to care and supplement traditional treatment. At the same time, leading reviews emphasize important limitations—including small samples, lack of rigorous active control conditions, inconsistent outcome measures, and limited assessment of long-term effects, as well as inconsistency in how conversational outcomes and relational metrics are evaluated across studies (Bodner et al., 2026; Gallegos et al., 2024; Gong et al., 2026). In other words, while there is reason for cautious interest, the evidence base does not yet substantiate claims that AI can deliver therapeutic equivalence to human relational care—even where symptom change is observed—and it certainly does not address whether these effects are driven by relational processes akin to those identified as central in therapy (Flückiger et al., 2018; Howcroft et al., 2026).

In the next part of this series, I will explore what therapy requires beyond mirrored reflection—what happens when we move from resonance to reciprocity, and why human subjectivity cannot be fully replaced by even the most sophisticated simulated empathy.

References

Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of attachment: A psychological study of the strange situation. Erlbaum.

Benjamin, J. (2004). Beyond doer and done to: An intersubjective view of thirdness. The Psychoanalytic Quarterly, 73(1), 5–46. https://doi.org/10.1002/j.2167-4086.2004.tb00151.x

Benjamin, J. (2017). Beyond doer and done to: Recognition theory, intersubjectivity and the third. Routledge.

Bodner, R., Lim, K., Schneider, R., & Torous, J. (2026). Efficacy and risks of artificial intelligence chatbots for anxiety and depression: A narrative review of recent clinical studies. Current Opinion in Psychiatry, 39(1), 19–25. https://doi.org/10.1097/YCO.0000000000001048

Bowlby, J. (1982). Attachment and loss: Vol. 1. Attachment (2nd ed.). Basic Books. (Original work published 1969)

Bruner, J. (1990). Acts of meaning. Harvard University Press.

Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316–340. https://doi.org/10.1037/pst0000172

Feng, X., Tian, L., Ho, G. W. K., Yorke, J., & Hui, V. (2025). The effectiveness of AI chatbots in alleviating mental distress and promoting health behaviors among adolescents and young adults: A systematic review and meta-analysis. Journal of Medical Internet Research, 27, e79850. https://doi.org/10.2196/79850

Gallegos, D., Chen, E., & Liu, H. (2024). Limitations in AI chatbot mental health interventions: A systematic review. Journal of Technology in Behavioral Science, 9(2), 105–123. https://doi.org/10.1007/s41347-024-00321-5

Gong, J., Wen, X., Tao, F., Wang, X., Yang, X., & Tang, Y. (2026). Evaluating text‑based conversational agents for mental health: A systematic review of metrics, methods, and usage contexts. arXiv. https://doi.org/10.48550/arXiv.2602.17669

Howcroft, A., Bennett‑Weston, A., Khan, A., Griffiths, J., Gay, S., & Howick, J. (2026). AI chatbots versus human healthcare professionals: A systematic review and meta-analysis of empathy in patient care. arXiv. https://doi.org/10.48550/arXiv.2602.05628

Hull, T. D., Zhang, L., Areán, P. A., & Malgaroli, M. (2025). Mental Health Generative AI is safe, promotes social health, and reduces depression and anxiety: Real‑world evidence from a naturalistic cohort (Preprint). arXiv. https://doi.org/10.48550/arXiv.2511.11689

Li, Q., Zhang, Y., & Smith, J. (2023). Conversational agents for mental health: Mechanisms of change and effectiveness. Frontiers in Digital Health, 5, 1156789. https://doi.org/10.3389/fdgth.2023.1156789

Ma, N., Khynevych, R., Hao, Y., & Wang, Y. (2025). Effect of anthropomorphism and perceived intelligence in chatbot avatars of visual design on user experience. Frontiers in Computer Science, 7, Article 1531976. https://doi.org/10.3389/fcomp.2025.1531976

Nyakhar, S., & Wang, H. (2025). Effectiveness of artificial intelligence chatbots on mental health and well‑being in college students: A rapid systematic review. Frontiers in Psychiatry, 16, 1621768. https://doi.org/10.3389/fpsyt.2025.1621768

Rahsepar Meadi, M., Sillekens, T., Metselaar, S., van Balkom, A., Bernstein, J., & Batelaan, N. (2025). Exploring the ethical challenges of conversational AI in mental health care: Scoping review. JMIR Mental Health, 12, e60432. https://doi.org/10.2196/60432

Shu, C., Lai, K., & He, L. (2026). Human-AI attachment: How humans develop intimate relationships with AI. Frontiers in Psychology, 17, 1723503. https://doi.org/10.3389/fpsyg.2026.1723503

Sohn, J.-S., Ha, B.-G., Park, S., Kim, J.-J., Lee, E., Oh, H., … Kim, E. (2026). Systematic review and meta-analysis of chatbots in the management of depressive and anxiety symptoms. npj Digital Medicine. Advance online publication. https://doi.org/10.1038/s41746-026-02566-w

Truong, T. T. H., & Chen, J. S. (2024). When empathy is enhanced by human–AI interaction. Asia Pacific Journal of Marketing and Logistics. Advance online publication. https://doi.org/10.1108/APJML-11-2023-1385

Winnicott, D. W. (1971). Transitional objects and transitional phenomena. In Playing and reality (pp. 1–25). Routledge. (Original work published 1953)

Zhong, W., Luo, J., & Zhang, H. (2024). The therapeutic effectiveness of artificial intelligence-based chatbots in alleviation of depressive and anxiety symptoms in short-course treatments: A systematic review and meta-analysis. Journal of Affective Disorders, 356, 459–469. https://doi.org/10.1016/j.jad.2024.04.057