Background Socially interactive AI can elicit meaningful relational experiences. Studies often conflate attachment, anthropomorphism, companionship, dependence, and addiction, and group embodied robots with dissimilar software. Methods This critical narrative review of a reconstructed, non-exhaustive corpus ( Grant & Booth, 2009 ) is interpretive, not aggregative, because the initial search records disagree. We rebuilt the evidence set from the initially identified literature, separated primary from contextual sources, and sorted them by embodiment and dialogue architecture. Coding kept attachment functions distinct from related constructs. A reproducible OpenAlex and PubMed search (2010–2025) retrieved only one of the 12 primary reports. A coverage probe found 34 further provisionally eligible reports. We claim no exhaustive coverage or full PRISMA compliance. Results Reclassification left 12 primary reports and seven contextual sources. In these reports, relational significance appeared with embodied robots, non-LLM systems, and generative systems, carried by co-presence and touch or by conversational contingency, personalisation, and apparent continuity. Most measured companionship, trust, social presence, perceived support, or one problematic-use indicator. Few covered more than one attachment-system function. None established a behavioural-addiction syndrome. The predominantly cross-sectional, self-selected evidence cannot establish whether heavier use causes loneliness. Conclusions A relationship with an AI system can matter without amounting to reciprocal human attachment or behavioural addiction. Within the reviewed corpus, the evidence supports a broad, uneven continuum of involvement. Omitted studies may change the apparent mix of constructs, technologies, and outcomes. Progress requires longitudinal, technology-stratified studies with construct-specific measures that establish temporal order and impairment and separate embodiment from generative capability.
No takes yet. Share an insight, caveat, or question.
IMPRAM et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: