The Heterogeneous Effects of AI Companionship: An Empirical Model of Chatbot Usage and Loneliness and a Typology of User Archetypes

Auren R. Liu, Pat Pataranutaporn, Pattie MaesAAAI 2025 · DOI

Abstract

Companion chatbots offer a potential solution to the growing epidemic of loneliness, but their impact on users' psychosocial well-being remains poorly understood, raising critical ethical questions about their deployment and design. This study presents a large-scale survey (n = 404) of regular users of companion chatbots, investigating the relationship between chatbot usage and loneliness. We develop a model explaining approximately 50% of variance in loneliness; while usage does not directly predict loneliness, we identify factors including neuroticism, social network size, and problematic use. Through cluster analysis and mixed-methods thematic analysis combining manual coding with automated theme extraction, we identify seven distinct user profiles demonstrating that companion chatbots can either enhance or potentially harm psychological well-being depending on user characteristics. Different usage patterns can lead to markedly different outcomes, with some users experiencing enhanced social confidence while others risk further isolation. These findings have significant implications for responsible AI development, suggesting that one-size-fits-all approaches to AI companionship may be ethically problematic. Our work contributes to the ongoing dialogue about the role of AI in social and emotional support, offering insights for developing more targeted and ethical approaches to AI companionship that complement rather than replace human connections.

Introduction

As the global "loneliness epidemic" intensifies, millions of individuals are turning to AI companions like Replika, Character.ai, and Snapchat MyAI for emotional support. This shift has sparked a divide in the scientific community: is AI a "digital band-aid" that fosters social confidence, or a "toxic substitute" that encourages withdrawal from real-world relationships? This research moves beyond the binary question of whether AI helps or hurts, instead investigating for whom these technologies are beneficial and for whom they pose risks. By examining the interplay between personality traits, usage patterns, and social context, the study seeks to provide a granular understanding of AI companionship.

Methods

The researchers employed a mixed-methods approach to capture both quantitative trends and qualitative nuances. A large-scale survey was conducted with 404 regular users (defined as those using chatbots at least weekly for over a month). Participants were assessed using established psychometric scales, including the UCLA Loneliness Scale (ULS-8), the Big Five Inventory (BFI-10) for personality, and the Generalized Problematic Internet Use Scale (GPIUS2).

Statistical analysis involved multiple regression to identify predictors of loneliness and K-means clustering to categorize users into distinct archetypes. To ensure depth in understanding user motivations, the team utilized an AI-augmented thematic analysis, combining manual coding of open-ended responses with GPT-4-based topic extraction.

The User Base: A Profile of AI Companionship

The study focused on a specific demographic of "power users" who have integrated AI into their daily lives over significant periods. Unlike studies using general populations, this dataset captures the behavior of individuals for whom AI companionship is a regular fixture.

Distribution of chatbot usage patterns among survey participants, highlighting the high frequency and established nature of usage in the sample.

Demographically, the sample was primarily aged 30–35, 93.6% from the US, and majority male (59.7%). The most common platforms were Snapchat MyAI (31.6%), Character.ai (30.4%), and Replika (22.6%). As shown in the usage data, a significant portion of the sample engages with chatbots multiple times per week, often in sessions lasting 15 to 30 minutes.

Empirical Findings: Predictors of Loneliness

A central finding of the research is that simply spending time with a chatbot does not correlate directly with loneliness. Instead, the relationship is mediated by psychological traits and specific usage patterns. The researchers developed a regression model explaining approximately 50% of the variance in user loneliness.

Correlation matrix illustrating the relationships between main variables of interest, participant characteristics, and perceptions of the chatbot.

Key predictors identified include:

  • Neuroticism (β = 0.358): The strongest personality-based predictor of high loneliness.
  • Problematic Chatbot Use (β = 0.166): Addiction-like symptoms and compulsive usage patterns were positively correlated with loneliness.
  • Social Network Size (β = -0.316): This remained the strongest protective factor against loneliness.

Interestingly, for individuals who highly value human connection, longer AI sessions were associated with increased feelings of loneliness. This suggests a "mismatch" where the AI fails to meet the high social standards of the user, potentially exacerbating feelings of isolation.

A Typology of User Archetypes

The study identified seven distinct user archetypes, demonstrating the heterogeneous impact of AI companionship:

  1. Well-Adjusted Moderate Users (23%): The largest group; they use AI for emotional processing while maintaining strong human ties.
  2. Fulfilled Dependent Users (18%): Characterized by high usage but low loneliness; they view the AI as a creative collaborator or beneficial supplement.
  3. Lonely Moderate/Light Users: These groups show signs of social withdrawal and use AI to fill a void they feel unable to fill with humans.
  4. Socially Challenged Frequent Users: High frequency but short sessions; they often struggle with trust in both humans and AI.

Discussion and Limitations

The findings suggest that AI companionship is not a monolithic experience. While some users find it a helpful tool for social rehearsal and emotional support, others risk further isolation through problematic use.

However, the study's cross-sectional nature means it cannot definitively prove causality—whether chatbots cause loneliness or if lonely individuals are simply more drawn to them. Furthermore, the sample was skewed toward Western, white, male-identifying participants. Future work should involve longitudinal studies to track how these archetypes evolve and the development of adaptive interfaces that can detect "at-risk" usage patterns to suggest real-world social interventions.

Summary updated

Abstract

Companion chatbots offer a potential solution to the growing epidemic of loneliness, but their impact on users' psychosocial well-being remains poorly understood, raising critical ethical questions about their deployment and design. This study presents a large-scale survey (n = 404) of regular users of companion chatbots, investigating the relationship between chatbot usage and loneliness. We develop a model explaining approximately 50% of variance in loneliness; while usage does not directly predict loneliness, we identify factors including neuroticism, social network size, and problematic use. Through cluster analysis and mixed-methods thematic analysis combining manual coding with automated theme extraction, we identify seven distinct user profiles demonstrating that companion chatbots can either enhance or potentially harm psychological well-being depending on user characteristics. Different usage patterns can lead to markedly different outcomes, with some users experiencing enhanced social confidence while others risk further isolation. These findings have significant implications for responsible AI development, suggesting that one-size-fits-all approaches to AI companionship may be ethically problematic. Our work contributes to the ongoing dialogue about the role of AI in social and emotional support, offering insights for developing more targeted and ethical approaches to AI companionship that complement rather than replace human connections.
BibTeX
@inproceedings{liu2025heterogeneous,
  author = {Auren R. Liu and Pat Pataranutaporn and Pattie Maes},
  title = {{The Heterogeneous Effects of AI Companionship: An Empirical Model of Chatbot Usage and Loneliness and a Typology of User Archetypes}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.1609/aies.v8i2.36658},
}
Figure