Trust, Attachment, and Behavioral Engagement in Human–AI Interaction: A Systematic Review and Preliminary Meta-Analysis

Abstract

The interaction between humans and Artificial Intelligence encompasses three distinct domains; social, advisory, and companionship. When studying these domains, it is essential to recognize the importance of psychological concepts like trust, closeness and engagement. Even though separate domains have been studied separately, there is research yes that competency assessments, social signs, feelings and dependency might be interconnected.

The purpose of this systematic review and preliminary meta-analysis is to synthesize the quantitative and qualitative literature that links human-artificial intelligence (AI) interaction variables to user trust, attachment, and behavior engagement. Variables examined include anthropomorphism, perceived competency, social presence, loneliness, and interaction intensity.

The provided review report included search results from PsycINFO, Scopus, Web of Science, ACM Digital Library, IEEE Xplore, and Google Scholar up until June 2025. Following subsequent bibliographic verification, 8 research reports were identified as the confirmed core evidence base for this review. Since the extracted record did not provide sufficient independent, comparable effect sizes, no pooled coefficient could be calculated. Descriptive synthesis was used to summarize the quantitative findings while thematic analysis was conducted on the qualitative evidence.

Results: Experimental and observational proof indicated that competence, dependable overall performance, and selected social cues have been associated with trust, at the same time as anthropomorphic consequences trusted cue feature and undertaking context. Attachment evidence became sparse and heterogeneous. Qualitative research related trust, self-disclosure, responsiveness, social motivation, and interplay depth with dating development. Behavioral engagement and reliance had been related to agree with and social interaction throughout numerous operationalizations.

Conclusions: The proof supports a initial twin-pathway framework distinguishing cognitive consider from affective attachment. These pathways can also converge on behavioral engagement, however the gift literature does no longer establish causal mediation or a common quantitative effect. A preregistered, construct-harmonized meta-evaluation is warranted.