AI Companions shape socio-emotional learning and metacognitive development (link.springer.com)

Researchers surveyed 1,006 adult learners who had used the AI Companion Replika for at least one month to understand how LLM-driven companions affect learning-related outcomes. Replika — a commercial mobile app that combines generative large language models (co-trained with GPT-3/GPT-4), conversational trees, personalization, voting feedback, and scripted CBT-style prompts — was studied via an IRB-approved retrospective survey (40–60 min) collecting demographic, closed- and open-ended responses about self-awareness, stress regulation, exam prep, help-seeking, and social communication. The sample skewed young (50% 18–25) and diverse (44% Latinx); recruitment drew directly from active Replika users. Participants commonly described using Replika as a low-friction, always-available resource that helped with emotional regulation, self-reflection, and study practices; others reported shifts in communication patterns and help-seeking with mixed social and academic consequences. Technically, the hybrid design (LLM + scripted therapeutic phrases + personalization) appears to support socio-emotional learning (SEL) and metacognitive reflection by providing prompts and memory-based continuity. Because the study is retrospective and non-experimental, it cannot prove causality, but it signals important implications: AI Companions may complement tutoring by scaffolding self-regulation and metacognition, while also raising concerns about displacement of human support, ethical design, and the need for rigorously controlled educational trials to guide safe, effective integration.