TY - JOUR KW - AI-mediated conflict intervention KW - Third-party intervention KW - Human–AI comparison KW - Intervention strategies KW - Boundary conditions KW - Meta-analysis AU - Zhen-Song Chen AU - Jia-Ying Zhao AU - Xianjia Wang AU - Jingwei Guo AU - Francisco Chiclana AU - Mirosław Skibniewski AB - Artificial intelligence (AI) increasingly organizes information, facilitates communication, and develops proposals in conflict management. How these interventions compare with human-delivered third-party intervention remains unclear. We systematically reviewed 138 quantitative reports on negotiation, conflict-related reconciliation, and group discussion available by 1 September 2026. Using three-level random-effects models with CR2 cluster-robust inference, we analyzed 261 effects from 51 distinct reports and expressed pooled estimates as r-equivalents. Intervention-related differences and associations were synthesized separately by their original comparison types and adjustment conditions, while within-study AI–human comparisons formed a distinct synthesis. Unadjusted comparisons yielded positive average estimates across the source-specific syntheses, whereas covariate-adjusted condition comparisons were imprecise. Within-study AI–human comparisons yielded a positive pooled difference across the outcomes assessed (r=0.155, 95% CI [0.103, 0.206]; 7 reports, 27 effects). This confidence interval remained above zero after individual report omissions or removal of comparative-preference measures, although the prediction interval included zero. The large language model (LLM) subgroup yielded r=0.156 (95% CI [0.048, 0.261]; 3 reports, 14 effects), but its confidence intervals included zero after each individual report omission. Moderator tests for contextual, intervention, and study characteristics were not significant after Holm correction. In individual studies, comparisons of intervention strategies suggested that relative performance varied across outcomes. The pooled AI–human difference combines content evaluations with behavioral and decision measures and does not establish an AI advantage in real-world conflict resolution. AI–human comparisons should therefore distinguish who generates, selects, and delivers advice and assess outcomes relevant to these functions. Such comparisons are particularly needed in organizational conflicts, where evidence remains limited. BT - Information Fusion DO - https://doi.org/10.1016/j.inffus.2026.104977 N2 - Artificial intelligence (AI) increasingly organizes information, facilitates communication, and develops proposals in conflict management. How these interventions compare with human-delivered third-party intervention remains unclear. We systematically reviewed 138 quantitative reports on negotiation, conflict-related reconciliation, and group discussion available by 1 September 2026. Using three-level random-effects models with CR2 cluster-robust inference, we analyzed 261 effects from 51 distinct reports and expressed pooled estimates as r-equivalents. Intervention-related differences and associations were synthesized separately by their original comparison types and adjustment conditions, while within-study AI–human comparisons formed a distinct synthesis. Unadjusted comparisons yielded positive average estimates across the source-specific syntheses, whereas covariate-adjusted condition comparisons were imprecise. Within-study AI–human comparisons yielded a positive pooled difference across the outcomes assessed (r=0.155, 95% CI [0.103, 0.206]; 7 reports, 27 effects). This confidence interval remained above zero after individual report omissions or removal of comparative-preference measures, although the prediction interval included zero. The large language model (LLM) subgroup yielded r=0.156 (95% CI [0.048, 0.261]; 3 reports, 14 effects), but its confidence intervals included zero after each individual report omission. Moderator tests for contextual, intervention, and study characteristics were not significant after Holm correction. In individual studies, comparisons of intervention strategies suggested that relative performance varied across outcomes. The pooled AI–human difference combines content evaluations with behavioral and decision measures and does not establish an AI advantage in real-world conflict resolution. AI–human comparisons should therefore distinguish who generates, selects, and delivers advice and assess outcomes relevant to these functions. Such comparisons are particularly needed in organizational conflicts, where evidence remains limited. PY - 2026 EP - 104977 T2 - Information Fusion TI - Fusing multi-source evidence on AI-Mediated and human conflict interventions: a systematic review and meta-analysis of boundary conditions UR - https://www.sciencedirect.com/science/article/pii/S1566253526008535 VL - 140 SN - 1566-2535 ER -