Loneliness is increasingly recognised as a public health concern with documented psychological and
physiological costs. Generative AI systems now offer “AI companions” capable of sustained, responsive
conversation. This article explores emerging evidence on their psychological effects. Short-term
reductions in loneliness have been observed following chatbot interactions, with perceived responsiveness
(“feeling heard”) implicated as a key mechanism. However, outcomes are not uniform: multi-method
and longitudinal data indicate that high-intensity, companionship-oriented use - particularly in the
context of limited offline support - is associated with higher loneliness, reduced time spent with people,
and greater emotional dependence. Integrating findings from psychology and social neuroscience, the
article traces (i) how AI companions may confer benefit through perceived social support, emotion
regulation, and felt understanding; (ii) how they may cause harm via over reliance, avoidance and
safety behaviours, and parasocial attachment; and (iii) the moderators - individual differences, usage
intensity, modality, and social context - that tilt outcomes either way. AI companions are treated as
potential adjuncts to human ties rather than substitutes.
