ChatGPT tries to understand the patient. What it really can - and can't - do in CBT case conceptualization
Case conceptualization is the heart of CBT therapist work. It’s the moment when the therapist stops collecting facts and begins to understand - why this patient, in this situation, reacts this way. It’s a process requiring both theoretical knowledge and clinical intuition built over years.
Can AI help with this?
Researchers from Sigmund Freud University in Milan decided to find out. Their paper, published in 2026 in Cyberpsychology: Journal of Psychosocial Research on Cyberspace, is one of the first studies analyzing ChatGPT-4’s capabilities directly in the context of case conceptualization - not in psychotherapy generally, but in this most demanding, most clinical aspect of therapeutic work.
What was studied and how
The authors (Buattini, Barjami, Paponetti et al.) applied the LIBET conceptualization model - a cognitive therapy tool that structures patient understanding through two key constructs: life themes (deep beliefs and emotional patterns) and semi-adaptive plans (strategies the patient developed to cope with difficulties - not necessarily effective, but seemingly logical).
Ten participants went through semi-structured interviews. Transcriptions were anonymized and provided to ChatGPT-4 with detailed LIBET model instructions. Results were subjected to reflexive thematic analysis - a qualitative method capturing both the system’s strengths and recurring errors.
What AI did well
The results are not straightforward - and that’s exactly what makes them valuable. ChatGPT-4 demonstrated several real strengths.
Consistent structure and material organization. The system consistently organized interview information, applied proper LIBET terminology, and generated responses in a clear, logical layout. For a therapist who sits down to write notes after a long day of sessions, such structural assistance has real value.
Hypothesis-based reasoning with cautious language. ChatGPT didn’t declare - it proposed. It formulated hypotheses, used conditional language, and flagged uncertainty. This matters: the system behaved in line with what good CBT conceptualization should look like - as a proposal for verification, not a diagnosis.
Partial reproduction of LIBET logic. The model recognized and applied theoretical constructs, suggesting that with properly constructed instructions, AI can operate within a specific therapeutic approach - not just as a general text generator.
Where AI stumbled
The study is honest in describing limitations - and this is its greatest value for CBT practitioners.
Confusing strategies with vulnerabilities. The most common error was that the system confused semi-adaptive plans (coping strategies) with life themes (deeper emotional patterns). This is exactly the kind of mistake that in real therapy can lead to working at too superficial a level - one that an experienced therapist catches intuitively, but AI - not yet.
Difficulties with relational abstraction. Life themes are constructs requiring understanding of context, history, and ambiguity. ChatGPT performed significantly better with what’s concrete and structured than with what’s deeper and more relational.
Interpretation without verification. The system sometimes generated interpretations that sounded convincing but weren’t grounded in interview content. This phenomenon known as hallucination takes on particular weight in a clinical context.
What this means for therapeutic practice
The authors formulate a cautious but specific conclusion: ChatGPT can be a useful tool for reflection - particularly in supervision and training - but is not ready for independent use in clinical reasoning and decision-making.
This is an important distinction. It’s not that AI isn’t suited for CBT. But AI can help organize material, generate hypotheses for verification, and structure notes - provided the therapist verifies every step.
For CBT therapists working with conceptualization, this is a practical question: how much time does it take you just to structure material collected from interviews and subsequent sessions? How much of that is mechanical work - organizing, naming, fitting into a framework - and how much is actual clinical reasoning?
Therapy Support perspective
At Therapy Support, we’re building a tool that addresses exactly this question. We don’t replace the therapist’s clinical reasoning. We support the part of work that is mechanical but essential: extracting information from transcriptions, structuring notes, organizing material according to CBT logic - as proposals for therapist verification.
The Buattini et al. study confirms that this boundary - between organizing material and clinical interpretation - is real and meaningful. An AI system can operate valuably on the first side. The clinical decision always belongs to the therapist.
Summary
This is the first study of its kind to pose such a specific question: can AI handle case conceptualization in CBT? The answer is: partially yes, but with significant caveats. Strengths - structure, hypothesis language, terminology - are promising. Weaknesses - difficulties with abstraction and risk of misinterpretation - remind us that therapist oversight remains necessary.
For CBT practitioners, it’s worth following this line of research. Not to entrust conceptualization to a machine - but to know where AI can genuinely lighten the load, and where it must not be trusted without verification.
Source: Buattini M., Barjami D., Paponetti L., Torres D., Borlimi R., Caselli G. Evaluating strengths, limitations, and future directions of ChatGPT in psychological analysis within case conceptualization: A qualitative analysis. Cyberpsychology: Journal of Psychosocial Research on Cyberspace, Vol. 20, No. 1, 2026. DOI: 10.5817/CP2026-1-4
Therapy Support is a tool supporting documentation and work organization for CBT psychotherapists. All proposals generated by the system are starting material for therapist verification and decision-making.