CBT & AIAdjunct LLM useAnxiety-depression disordersDoctoral student

Luc Garczynski

Doctoral student in psychology — Université de Montréal

In brief: Luc Garczynski is a clinical psychologist trained in CBT at the University of Strasbourg, currently a doctoral student at the Université de Montréal. His thesis project investigates the responsible integration of an LLM chatbot as an adjunct to CBT for adults presenting with anxiety-depression disorders. His approach is explicitly synergistic: he does not seek to replace the therapist with AI, but to map the conditions under which an LLM can strengthen between-session work and patient autonomy, under clinical supervision.

Profile

Institution: Université de Montréal — Department of Psychology

Status: Doctoral student in psychology (2025–2030), North American model with two preparatory years

Training: Master 2 in Cognitive and Behavioural Therapies (University of Strasbourg, 2024). Additional training in trauma-focused hypnosis.

Clinical practice: Psychologist specialised in CBT. Clinical placement at Psy Intégrative Montréal (2024–2025) under the supervision of Céline Castillo, with patients presenting PTSD, school phobias, or anxiety disorders (children, adolescents, veterans, victims).

Industry collaborations: Former consultant for Feel (mobile mental health app, design of ACT psychoeducational sessions).

A practitioner-researcher profile at a crossroads

What distinguishes Luc Garczynski is the explicit articulation of four entry points he refuses to silo:

CBT clinical practice

Training and practice in cognitive-behavioural therapies, with patients seen during clinical placement. A clinical anchor that keeps the research grounded in the constraints of practice.

Methodological research

Intensive training in PRISMA systematic reviews and advanced statistics, with three associated research studies conducted before doctoral work (Brief ACT, oncohaematology, ADHD neuropaediatrics).

Digital design

Experience as a designer of digital therapeutic content at Feel: he knows what it concretely means to “hold” a digital self-therapy device.

Personal AI use

Heavy LLM user since 2022, supporting his own theoretical and clinical reflection. The “from the inside” experiential dimension is claimed as a methodological prerequisite for any research on the tool.

This combination allows him to evaluate technological devices not from a purely theoretical or purely technical viewpoint, but from the joint experience of clinician, methodologist, and user.

Doctoral project

Provisional title: “Responsible integration of an LLM chatbot as adjunct to CBT for anxiety-depression disorders in adults.”

Research problem: anxiety and depressive disorders affect roughly 11.3% of adults, and the supply of CBT remains far too low to meet demand. A promising avenue is to integrate an LLM chatbot under clinical supervision to strengthen the between-session work, a central lever of CBT. But the concrete conditions and modalities of this integration remain poorly defined. Luc addresses this through a mixed and participatory approach.

Objective 1 — PRISMA-ScR scoping review

Mapping of adjunct LLM uses in adult CBT. Protocol filed on OSF, methodology compliant with PRISMA-ScR recommendations. Goal: build a minimal reference framework of integration configurations already documented in the literature, independent of specific technical choices (fine-tuning, RAG, framing). This first study structures the rest of the project.

Objective 2 — Co-construction of an operational framework

Participatory approach bringing together three groups of stakeholders: patients, CBT psychologists, AI developers. Mixed focus groups aimed at producing an operational integration guide covering tasks delegated to the LLM, the roles of each actor, clinical safeguards, escalation procedures in case of risk, and traceability of interactions.

Objective 3 — Randomised add-on trial

Clinical study comparing CBT alone vs CBT + LLM chatbot over 6 weeks, N=30 adults presenting mild to moderate anxiety-depression disorders. The aim is not to test an isolated chatbot, but to measure the incremental added value of an adjunct device structured by the operational framework built in Objective 2.

Cross-cutting objective — Web app prototype

Development of an operational prototype: LLM API + secure RAG + therapist-patient scripts. The tool is not the goal of the thesis, but the support for the Objective 3 trial. It embodies the project’s twofold requirement: concrete enough to be tested in ecological conditions, framed enough to remain clinically responsible.

Focus: the LLM-patient-therapist triad

Luc Garczynski refuses the binary opposition “autonomous AI vs traditional psychotherapy”. In his framework, the integration of AI in psychotherapy must be thought of as a triad: LLM, patient, therapist, where each pole carries its own parameters and where the three crossed interactions (LLM×patient, LLM×therapist, patient×therapist) must be mapped separately.

The three poles each have their parameters

The LLM (model type, training, safeguards), the patient (clinical profile, autonomous or framed use, attribution of success), the therapist (training, equipment, posture toward the tool). These variables are not interchangeable.

The three interactions are distinct

How the patient uses the LLM outside session is not the same question as how the therapist relies on the LLM to prepare their work, nor as how therapist and patient discuss what was said to the machine between two appointments.

Parameters are variables on a spectrum

A parameter, in Luc’s vocabulary, is “something that must vary across uses, on a spectrum”: degree of integration, degree of safeguard, type of safeguard. The scientific stake is to identify which parameters are clinically relevant, not to prescribe a single configuration.

This triadic framework usefully reads in dialogue with the taxonomy of Stade et al. 2024 (assistive, collaborative, substitutive integration).

What brought him to AI in psychotherapy

Luc’s interest in AI is not a fad. It begins with an older question: that of the large-scale dissemination of psychotherapy. Attentive since his bachelor’s degree to digital interventions in mental health, he sees in the arrival of LLM chatbots a qualitative turning point.

“I was struck by their capacity to support natural-language exchanges of great richness and to mobilise psychological content interactively. It is this evolution that progressively oriented my interest toward their possible place in the psychotherapeutic field.”

We will soon devote a longer testimonial to this trajectory (from the discovery of ChatGPT in M2 to the pivot toward AI in psychotherapy): we will return to it in detail in a dedicated interview.

Uses observed in the field

Luc’s observations, drawn from his own intern practice, his interviews with CBT psychologists, and the literature he is reviewing for his scoping review, sketch a contrasted landscape:

•
On the patient side: many patients already use ChatGPT or other LLMs spontaneously, as a support to think, put words on what they are going through, or extend reflections between sessions. Luc is starting to mobilise these tools in a framed way in some of his follow-ups.
•
On the therapist side: uses are numerous but concentrated on peripheral tasks: administrative work, writing, support to clinical reflection.
•
Central observation: there are still very few situations in which the LLM is integrated explicitly into therapy as a full-fledged therapeutic tool, mobilised in service of the change process. It is precisely this near-empty zone that his doctoral research aims to explore.

Anticipated potential for the future

Asked where LLMs will take the psychotherapy field, Luc’s answer is measured but clear.

“In my view, their main potential lies less in creating new therapy models than in their capacity to strengthen the effectiveness of existing therapy, by facilitating its implementation, addressing certain obstacles encountered by patients, and more actively supporting therapeutic processes between sessions.”

This stance translates into two practical anticipations: the continued growth of self-therapy uses with generic tools (which must be documented rather than declared dangerous), and the gradual emergence of framed clinical integrations, accompanied by the therapist, especially on the classic levers of CBT where between-session work plays a central role (cognitive restructuring, gradual exposure, psychoeducation, relaxation).

Publications and outputs

Adjunct patient use of large language models alongside psychotherapy: A scoping review (OSF protocol, in progress)

Scoping review (PRISMA-ScR methodology) mapping documented configurations of LLM use by patients in parallel with psychotherapy. This is the founding study of his doctoral thesis, with the protocol filed on the OSF platform.

Exploratory qualitative study — ChatGPT uses among CBT psychologists (Qualitative methods coursework, 2025)

Interpretive descriptive qualitative study interviewing four CBT psychologists about their uses of ChatGPT in clinical practice. Work conducted as part of a team project, of which Luc methodologically led the bulk of the process.

Systematic review — Brief ACT Interventions (Psy.link, supervision Linda Kempe, 2024–2025)

Associated research on the effectiveness of brief ACT interventions for anxiety-depression symptoms. PsycInfo/PubMed/Science Direct screening, critical analysis, editorial contribution.

Systematic review — Psychosocial needs in cell therapy (Université de Montréal, 2025)

PRISMA review on the psychosocial needs of patients undergoing cell therapy for haematological cancers. Double-coding via Covidence, data extraction, methodological quality assessment.

Quantitative study — Barkley Programme and ADHD siblings (Dr Sonja Finck, Strasbourg, 2023–2025)

Impact study of the Barkley Programme on the siblings of children with ADHD. Statistical processing of 21 questionnaires per family for 102 families. Mediation and moderation analyses.

Why this matters for clinicians

Luc Garczynski occupies a singular position among young researchers working on AI in psychotherapy. Four reasons to follow his work:

Synergy rather than substitution: his explicit axis is to strengthen existing therapy, not replace it. This stance avoids the false oppositions that clutter public debate.
Clear CBT anchoring: his work addresses a delimited clinical field (CBT for adult anxiety-depression disorders), allowing practitioners to know precisely which part of their activity is concerned.
Participatory method: by integrating from Objective 2 patients, psychologists, and developers, he builds a framework that will not be imposed on clinicians from the outside.
Practitioner and researcher: he continues to see patients during clinical placement throughout his thesis. The safeguards he describes are safeguards he himself must design for his own practice.

Related concepts on this site

Luc Garczynski’s work directly intersects several concepts documented in our fact sheets:

Epistemic double standard

AI evaluation in mental health tends to apply to technological devices requirements that are never applied to the human reference practices. A critical concept mobilised in reading the studies that Luc reviews.

WEIRD Sample

Most published studies on AI in mental health rely on WEIRD (Western, Educated, Industrialized, Rich, Democratic) samples, limiting the clinical generalisation of their conclusions.

Our collaboration

We have maintained an ongoing collaboration with Luc Garczynski since March 2026, articulated around several axes:

• Cross-contributions to the doctoral scoping review: reference sharing, methodological discussions, editorial feedback on successive versions.
• Series of co-signed articles for this site: popularising the conceptual framework of the thesis, the Stade 2024 framework, and the identified clinical applications.