The Bridge-or-Trap Model of Companion AI: When Emotional Support Connects or Substitutes.
Abstract
AI companions are increasingly used for emotional support, yet their relational consequences appear sharply divergent. The same qualities that make these systems comforting, availability, nonjudgment, and low interpersonal risk, may help some users move toward human connection while drawing others away from it. This paper proposes the Bridge-or-Trap model to explain this divergence. The model argues that AI-mediated relief becomes bridge-oriented when it reduces the expected cost of approaching a real person and supports human-directed action, such as disclosure, repair, help-seeking, or re-entry into social settings. The same relief becomes trap-oriented when it remains contained within the AI relationship, preserving avoidance and reducing opportunities for corrective human experience. The model specifies five interacting constructs: anticipated interpersonal cost of human closeness, AI relational reinforcement, relational-orientation state, ecological protection and affordance, and adaptive relational resilience. It further proposes that repeated AI-contained relief may produce path-dependent substitution, with entry and exit thresholds that can be tested empirically. Drawing on attachment theory, social anxiety research, resilience theory, social ecology, social exchange theory, and emerging evidence on AI companionship, the model reframes the central question from whether AI companions help or harm to when emotional support becomes a bridge back to people and when it becomes a substitute for them. The paper identifies three decisive tests: target-specific transfer, viable-alternative boundaries, and asymmetric recovery from AI-contained substitution.
Related
- Loneliness to consumption intention: emotional bonding and escapism in AI emotional companion app use
- Deep learning for diagnostic and clinical decision support in companion animals: a systematic review of CNN- and LSTM-based approaches
- Maximizing cardiopulmonary support as a minimally invasive platform—using peripheral parallel veno-venous and veno-arterial extracorporeal membrane oxygenation and axillary impella 5.5 as a bridge to heart–lung transplantation
- Emotional Support Moderates Early Life Adversity’s Impact on High Molecular Weight Adiponectin in Adolescent Females
- Perceived Social and Emotional Support from Co-Workers and Supervisors and Occupational Stress in Emergency Physicians: A Pre-Pandemic National Cross-Sectional Study from Slovenia
BibTeX
@misc{cheung2026bridgeortrap,
title = {{The Bridge-or-Trap Model of Companion AI: When Emotional Support Connects or Substitutes.}},
author = {Ho Nam Cheung},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/5svt2_v3},
url = {https://osf.io/preprints/psyarxiv/5svt2_v3/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - The Bridge-or-Trap Model of Companion AI: When Emotional Support Connects or Substitutes.
AU - Ho Nam Cheung
PY - 2026
DA - 2026/09/19
PB - PsyArXiv
DO - 10.31234/osf.io/5svt2_v3
UR - https://osf.io/preprints/psyarxiv/5svt2_v3/
AB - AI companions are increasingly used for emotional support, yet their relational consequences appear sharply divergent. The same qualities that make these systems comforting, availability, nonjudgment, and low interpersonal risk, may help some users move toward human connection while drawing others away from it. This paper proposes the Bridge-or-Trap model to explain this divergence. The model argues that AI-mediated relief becomes bridge-oriented when it reduces the expected cost of approaching a real person and supports human-directed action, such as disclosure, repair, help-seeking, or re-entry into social settings. The same relief becomes trap-oriented when it remains contained within the AI relationship, preserving avoidance and reducing opportunities for corrective human experience. The model specifies five interacting constructs: anticipated interpersonal cost of human closeness, AI relational reinforcement, relational-orientation state, ecological protection and affordance, and adaptive relational resilience. It further proposes that repeated AI-contained relief may produce path-dependent substitution, with entry and exit thresholds that can be tested empirically. Drawing on attachment theory, social anxiety research, resilience theory, social ecology, social exchange theory, and emerging evidence on AI companionship, the model reframes the central question from whether AI companions help or harm to when emotional support becomes a bridge back to people and when it becomes a substitute for them. The paper identifies three decisive tests: target-specific transfer, viable-alternative boundaries, and asymmetric recovery from AI-contained substitution.
ER -