Luc Garczynski, a young researcher in clinical psychology

Luc Garczynski is a CBT clinical psychologist and a PhD candidate at the University of Montreal. He took nearly three years between discovering ChatGPT and deciding to devote his thesis to it. Three years testing the tool, being disappointed, setting it aside, coming back to it, looking for counter-arguments in a class that didn’t answer, letting a doctoral project collapse before accepting another. The account of a trajectory where resistance to AI was not an obstacle but part of the work.

Luc Garczynski has been a PhD candidate in psychology at the University of Montreal since September 2025. Trained in cognitive and behavioral therapies at the University of Strasbourg, he is preparing a thesis on the responsible integration of an LLM chatbot as an adjunct to CBT for anxious-depressive adults. His positioning is explicitly synergistic: he defends neither the replacement of the therapist by AI, nor the maintenance of a practice impervious to the tool. But this position did not arrive all at once.

This testimony is the account of a trajectory. Not a discourse on method, not a theoretical position: the thread of concrete events that led a CBT student to tell himself it was time to stop pushing the question away and look it in the face.

In the third year: the seed you don’t see

The story begins before the encounter with LLMs. In his third year of the bachelor’s degree, Luc doesn’t yet know which master’s to head toward. He goes through the clinical psychology literature and comes across an article that stays with him. A digital intervention, designed by a research team around a certain Dr. Weiner, developed to support Chinese social workers during the SARS-CoV-1 outbreak. Encouraging results. Nothing specifically psychotherapeutic in the strict sense, but an idea that stays.

”I don’t know why it struck me so much. But I think I always had this idea: when I like something, I like to share it. And psychology is by far my first area of interest.”

What imprints itself then is not “AI is good.” It’s an older question: how do you spread psychotherapy on a larger scale? How do you reach people who will never set foot in a psychologist’s office? The question will run through the next five years, coming back regularly when Luc observes around him “since he was little, these adults who interact, his buddies, and tells himself: there are variables missing there in their reasoning, he’s suffering on points where he isn’t understood, he wouldn’t have gotten worked up there if he’d communicated differently.”

Master’s, winter 2022: the revelation by accident

The first encounter with ChatGPT is neither planned nor enthusiastic. It comes via a trivial detail: an English text too well written for its author.

”I was quite good at English, and my flatmate and his friend ask me to proofread their text, and I found it really well written. I was like: ah, but they copied it off the Internet, that’s a bit of a giveaway. I knew my friend’s level, I was like, he doesn’t write that. And he tells me, it’s the AI that did it. He shows me ChatGPT. And on top of that he says: yeah, but we don’t tell too many people, we have to keep it secret, otherwise it’ll blow up too much.”

Within two weeks, everyone was talking about it. For Luc, the first attempts are disappointing. He’s revising for his OCD exams, and he asks for a presentation of OCD treatment in CBT. The AI of the time (this is the first version of ChatGPT, GPT-3.5) returns something that impresses superficially. As soon as you dig, it degrades fast.

”I had hopes it would help me with the internship report, but it was a kind of verbal vomit, actually. I’ve always been too perfectionist to put out something like that.”

Ambivalence sets in. Phases of use, phases of abandonment. And, right away, in the background, an underlying anxiety: the potential of being replaced by AI. Luc describes it as an anthropological bias, “very striking from the start.” It’s the kind of anxiety you push away by changing the subject, by minimizing the tool, by decreeing that it’s too limited to be a threat.

The class that stirs things up: a whole room that doesn’t hear

A precise moment breaks the balance. During his CBT master’s, Luc takes a class given by Matthieu Ferry that discussed AI’s potential for radical transformation in our society and in our profession in particular. What strikes Luc isn’t only the content: it’s the gap with his classmates’ reactions.

”That class stirred me up for several days. As soon as it ended, I called my brother to talk about it. What struck me was the gap with my classmates’ reactions. Most relativized, some brushed it off with a ‘it’s like the computer, it’ll change our practices.’ But I sensed clearly, even talking with them afterward, that they weren’t measuring the scale of what was happening nor the speed at which it was going to accelerate.”

Luc describes this moment as a silent rupture: the others don’t see, he sees something but can’t yet name it. He keeps in mind the counter-arguments heard in class for months, confronting each argument with what he continues to read and observe. It is, in retrospect, the start of an in-depth work that will last three years.

At the time, this work stays confined in his head. What occupies his daily life as a researcher-in-training is a doctoral project in preparation with a startup (Feel) that was developing a mobile application for anxiety and depression. Luc works several hundred hours with their team, prepares to join their lab for his thesis, and sees concretely how you can put ACT psychoeducation into an app that fits in a pocket.

Master’s in Canada: AI becomes a tool for thinking

The master’s clinical internship takes place in Montreal, at the Psy Intégrative Montréal unit, under the supervision of Céline Castillo. Luc works with patients presenting PTSD, school phobias, anxiety disorders — children, adolescents, veterans, victims. It’s a dense internship, and for the first time Luc finds himself having to produce structured clinical reasoning daily, not only for academic assignments.

That’s where his personal use of AI changes in nature. It’s no longer about testing whether it can write an internship report: it’s about testing the tool as support for his own reflection, by cross-referencing what he reads, what he observes in the internship and what he formulates for himself.

”I told myself, actually, I can’t outsource. I have to understand as much as possible everything I do, and the AI is going to support that reflection. And when I’ve really laid down foundations, then the AI comes in and says: look at this, look at that. So right there, there’s an interaction on the level of work, of clinical reflection, etc.”

A rule establishes itself, almost without his knowing it: understand first, cross-reference second. Never recommend AI directly to a patient in his internship, but use it to strengthen his own reasoning before the session. It’s a stance he will find again, a few months later, when he interviews four CBT psychologists for his qualitative study: they too, without his prompting the answer, first used AI for themselves before considering a clinical use.

Summer 2025: Feel collapses, plan B becomes plan A

The summer of 2025 is a pivot. The startup Feel’s fundraising doesn’t go as planned. The doctoral plan with the Feel team collapses. Fortunately, Luc had “done his homework” by applying in parallel to the University of Montreal, with a thesis supervisor who worked on hypnosis in oncology and chronic pain.

”They were solid subjects, but they didn’t correspond to what I felt I had to go toward. I told him I wanted to work on something else. And he asked me: but what do you want to do?”

It’s in front of this question — “what do you want to do?” — that Luc realizes he can no longer afford the luxury of pushing away the question he “didn’t want to look at.”

The tipping point: “I’m going to take the plunge”

The decision is made without a watchword, without a detailed plan.

”There, I pick up my subject again, then finally I come to discuss it, and I tell myself: but actually, I’m going to take the plunge. I’m going to try to integrate AI into psychotherapy.”

Luc’s first line of thought is whether he’ll go for an autonomous format — an AI given instructions so that it does the therapeutic work itself. He digs into the literature. He sees that it overlaps heavily with the field of digital interventions. And he arrives at a semi-consensus that is also his own, deep down, since the master’s class that had stirred him: he will have to keep the best of both worlds, and speak of a synergy rather than a substitution.

The founding case: a teenager, a relaxation app

Meanwhile, a clinical case from his internship will act as a concrete trigger. Luc writes his master’s dissertation on a teenager in a situation of anxious school refusal. One of the elements of the protocol he sets up is relaxation training. He provides a guided-relaxation app and observes how the patient uses it.

”I watch how it works, how he played with it, and I tell myself: it’s crazy, the potential of this thing. I’d have wasted time getting him to train his breathing. Having this cognitive concentration of counting while he breathes, in an already-activated situation, rather than pulling out my sheet of paper that’s going to put him in a bubble… and there I’m done, but actually, that’s too powerful. And actually, AI can perfectly play that role.”

The observation is simple, but it overturns part of his theoretical questions: session time is precious, the skills to automate (guided breathing, thought restructuring, psychoeducation) are numerous, and the patient benefits from being able to work on them at their own pace, outside the session, with a support that never tires.

The experiential before the theory

On arriving in Montreal for his doctorate, Luc already has several years of personal AI use behind him. This detail is not an anecdote: it has become an asserted methodological prerequisite.

”Since I learned there was AI, I’ve used it a lot. And that’s also what I say in the article where I interview CBT psychologists. A point I note: they all started with using AI in their daily lives without necessarily even arriving at a very theoretical knowledge of how it works. The experiential relationship, where you see the flaws, you see the moments when it gets it wrong, the moments when it’s good, the moments when it’s interesting — I told myself that was a nice edge.”

It’s a discreet but strong stance. In a field where most criticisms (and much enthusiasm) come from researchers who have never really used LLMs other than in a public demonstration, Luc posits that direct experience is a minimal condition of rigor. It’s not about being “for” or “against” the tool, it’s about not speaking of an object you’ve never held in your hand.

Today: the triad and the patient mapping

Today, Luc’s doctoral project has taken shape. It is structured in three stages: a PRISMA-ScR scoping review to map the adjunct uses already documented, a participatory co-construction of an operational framework with three stakeholder groups (patients, CBT psychologists, AI developers), and a randomized controlled trial comparing CBT vs CBT + LLM chatbot over six weeks.

The conceptual framework he mobilizes is an LLM-patient-therapist triad: three poles, three crossed interactions, and a guiding idea — the clinically relevant parameters cannot be reduced to a single configuration. His question is not “should we integrate an LLM?” but how to distinguish the possible integration configurations from one another.

“What interests me here is just this articulation that a patient uses an LLM, whatever the modalities around it — whether it’s supervised, whether the LLM was fine-tuned, whether there’s a RAG. That’s what I want to study. How, what are the different configurations for integrating an LLM into psychotherapy?”

Luc defends the exploratory stance as a quality of his project, not a weakness. He knows his results will be imperfect, that they will prove partially false as the models evolve, that some of his parameters will be obsolete before the defense itself. His ambition is not to produce the perfect configuration: it’s to lay bricks that will make the next research easier.

What this trajectory teaches us

Three things, at least, worth pinning down.

Testimony collected in March and April 2026, over three interviews. Luc Garczynski is a PhD candidate at the University of Montreal, under the supervision of his thesis supervisor.