Are You Qualified, ChatGPT? Examining Clinical Skills and Competencies of ChatGPT in Delivering Systemic Interventions

Key Findings

• — 12 clinical supervisors (mean 18.5 years of clinical experience, all with at least five years of couple and family therapy supervision) rated the transcript; cultural competence M = 3.85 (SD 0.81, range 1-6), systemic family therapy skills M = 3.91 (SD 0.76, range 0-6) and integration of a systemic orientation into individual therapy M = 3.11 (SD 0.37, range 1-5), all described as moderate; therapist characteristics scored highest among common factors (M = 6.56, SD 1.04)
• — Only one of the 12 supervisors identified the therapist as an AI; most of the remainder predicted a therapist aged 26-35 (n = 8) with up to 500 hours of client contact (n = 10)
• — Named weaknesses were pacing, joining, and attending to cultural and contextual factors
• — The authors report 'objective misalignment', defined as conflict between their custom instructions and the model's learned engagement behaviour: when the two conflicted, the model reverted to its general behaviour because the configuration sat inside the provider's own system instructions

Methodology Notes

Journal of Marital and Family Therapy 2026;52(4):e70168, published online 2026-08-31 (Crossref published-online; print issue October 2026), CC BY 4.0, PMID 42671046, PMCID PMC13528278. Authors at Koç University and Istanbul Bilgi University (Türkiye) and Texas A&M University. n = 1 client (a university student outside psychology in a 2.5-year relationship, with three prior individual therapy sessions), n = 1 session, n = 1 transcript; psychiatric diagnosis and current intimate-partner violence were exclusion criteria, so the case is low-acuity relational rather than crisis. Supervisors were a convenience sample recruited by word of mouth (17 invited, 12 completed). The reported percentages for supervisor predictions (75.1%, 83.4%) do not divide cleanly by 12; the counts are the safer citation. Wiley blocks automated fetchers; verified via the PMC full text (eutils efetch, read) plus PubMed and Crossref.

Sources