Text

Large language models (LLMs) are increasingly being researched in therapeutic contexts. This exploratory mixed-methods study examines the interaction between participants and an LLM (ChatGPT 4.5) in the role of an art therapist compared to human-human interaction. To compare these experiences, this study focusses on the factors of perceived empathy, the feeling of being understood, and the experience of reflecting on the creative process/artwork. Two groups were compared (June/July 2025; individual online session with a human art therapist, N = 10 vs. individual LLM setting, N = 24) by having participants complete a questionnaire with open- and closed-ended questions after each interaction. Two participants who completed both settings were interviewed about their experiences using a semi-structured format. The findings indicate that, under the conditions of this study, it can be assumed that LLMs are not as good at conveying the feeling of being empathetically understood and perceived as human art therapists, with greater variability of ratings in the ChatGPT group. The image as a therapeutic third party changes its role. It becomes a purely analytical object, losing its function as a medium for shared, processual exploration of subjective meanings and feelings. Due to the small and unequal sample size and the non-standardized measurement instrument, optimization of the study design is advisable. Further research on human-AI interactions in an art therapy setting is recommended.