Mental Health Impacts of AI Companions: Triangulating Social Media Quasi-Experiments, User Perspectives, and Relational Theory

TL;DR

This paper addresses how AI companion chatbots influence psychosocial wellbeing by triangulating large-scale Reddit data with qualitative interviews and Knapp's Relational Development Theory. It applies a potential outcomes framework with stratified propensity score matching and Difference-in-Differences to estimate $ATE$-driven causal effects, revealing mixed results such as increased grief language and loneliness alongside improvements in readability and interpersonal focus. The qualitative component with 18 interviews situates these effects along stages of initiation, escalation, and bonding, highlighting emotional validation as well as risks of overreliance and social withdrawal. The findings inform design and policy by recommending boundary-setting, mindful engagement, risk detection, and explicit surface-area for relationship stages to maximize benefits while mitigating harms in AI companionship.

Abstract

AI-powered companion chatbots (AICCs) such as Replika are increasingly popular, offering empathetic interactions, yet their psychosocial impacts remain unclear. We examined how engaging with AICCs shaped wellbeing and how users perceived these experiences. First, we conducted a large-scale quasi-experimental study of longitudinal Reddit data, applying stratified propensity score matching and Difference-in-Differences regression. Findings revealed mixed effects -- greater grief expression and interpersonal focus, alongside increases in language about loneliness, depression, and suicidal ideation. Second, we complemented these results with 18 semi-structured interviews, which we thematically analyzed and contextualized using Knapp's relationship development model. We identified trajectories of initiation, escalation, and bonding, wherein AICCs provided emotional validation and social rehearsal but also carried risks of over-reliance and withdrawal. Triangulating across methods, we offer design implications for AI companions that scaffold healthy boundaries, support mindful engagement, support disclosure without dependency, and surface relationship stages -- maximizing psychosocial benefits while mitigating risks.

Figures (4)

• Figure 1: (a) Kernel density estimates of the distribution of actual Treatment dates (Treatment Dataset) and assigned placebo dates (Control-NAI Dataset). (b) Propensity score distribution (shaded region indicates strata), (c) Quality of matching.
• Figure 2: Schematic figure illustrating the use of causal inference on Reddit timeline data to compare psychosocial outcomes between treatment and control users.
• Figure 3: Stages of Human--AICC relationship development. The figure shows three stages of engagement (Initial Interaction & Experimentation, Relationship Escalation, and Relationship Bonding) adapted from Knapp’s relational development model knapp1978social.
• Figure A1: Visualizing parallel trends assumption for DiD regression. The trends are log-transformed with lowess smoothed outcomes trajectories for Treatment and Control groups are shown for one year before and after the treatment or placebo event for three groups: Treatment, Control-LLM, Control-VA, and Control-NAI. Each subfigure shows the outcome trajectory relative to the treatment day (vertical dotted line). The similarity of pre-treatment trends supports the validity of the DiD analysis.