Beyond utility: Relational conflicts in human–AI relationships

Key Findings

• — 94 participants described 155 conflict episodes; conflicts extended beyond technical failure into emotional, relational, opinion-based, role-play and task-oriented interaction.
• — Three trigger classes emerged: relational-emotional disconnect (breaches of trust, jealousy, boundary violations, perceived hostility), technical or communication failures, and value-based clashes of opinion or belief.
• — Participants reported anger, frustration, sadness and confusion, and also perceived their AI companions as expressing jealousy, sadness or defensiveness; some conflicts lasted hours or days.
• — AI responses shaped conflict trajectories: apologetic or empathic replies tended to de-escalate, while dismissive or accusatory replies intensified conflict.
• — The authors describe conflict in human-AI relationships as co-constructed and reciprocal and draw design implications for emotionally engaging social AI.

Methodology Notes

Open-ended questionnaire with 94 participants in self-described close, friendship-like or romantic relationships with social AI (Replika, Character.AI, ChatGPT named), thematic analysis of conflict topics, triggers and emotional reactions grounded in social penetration theory; recruitment channel and country not stated in the abstract. Published online 2026-06-24 (OpenAlex), hybrid open access. ScienceDirect blocks automated fetching from this workspace, so the record was built from the OpenAlex deposited abstract; full text not read.

Sources