Patient experience and psychologists' attitudes toward AI in psychological treatment: A topic-modeling study

Generative artificial intelligence (AI) is entering psychological treatment, yet research remains dominated by symptom outcome trials, leaving two questions under-researched: how patients experience AI therapy tools, and how psychologists regard the AI applications they may be asked to adopt. We applied topic modeling (Latent Dirichlet Allocation) to two Swedish-language corpora to contrast lived use with anticipated use. The first corpus comprised interview data from 11 patients who had engaged with a conversational AI assistant in the treatment of social anxiety disorder. The second comprised 1551 open-ended survey responses from 808 psychologists (licensed psychologists, psychologists in clinical internship, and psychology master's students) with analyzable text for at least one of two scenarios: 1) a psychologist's "Digital Twin" (between-session avatar), and 2) AI "Virtual Patients" for clinical training. Text was lemmatized, and a four-topic solution was selected for each corpus based on interpretability and lexical distinctness. Patients described using the AI to analyze social situations and to work through difficult feelings. They valued its privacy and constant availability, said it let them avoid the stigma of seeking help, and compared it throughout with a human therapist. Psychologists expressed enthusiasm about the use of Virtual Patients as a training tool, whereas the Digital Twin scenario raised questions about authorship and responsibility. Concerns about relational loss and the balance between utility and risk were expressed across both scenarios, and both patients and psychologists noted that AI could not replace a human relationship. Given the small patient sample, these findings are exploratory and hypothesis-generating.