AI as a Group Mediator: A Conceptual Framework for Triadic Chat-Based Therapy

At the intersection of artificial intelligence and systemic therapy lies a new possibility: triadic chat-based mediation, where a Large Language Model (LLM) participates as conversational facilitator between relational partners. This paper develops a conceptual framework outlining potential benefits and risks. Advantages may include continuous availability, enhanced memory for relational dynamics, reduced barriers to engagement, and support for balanced participation. Concerns arise regarding dependency, reduced self-regulation, hallucinations, bias, and risks of re-traumatization from persistent recall. The framework situates the LLM not as therapist but as tentative relational scaffold-a structure that might support dialogue but could equally risk interfering with natural conflict resolution if misapplied. By integrating perspectives from AI ethics, digital health regulation, and psychotherapy practice, the paper defines a research agenda for evaluating safety, clinical utility, and cultural fairness, advancing critical dialogue on AI’s emerging role in systemic therapy.