A psychotherapist uses AI for his own trauma, but would not prescribe it to his patients

By Zendoric · September 24, 2026.

Joe Nucci is a licensed psychotherapist (LPC, LMHC) and the author of a book, Psychobabble, with entire chapters devoted to warning against self-diagnosis. This year, after being assaulted, he ignored his own advice: during a sleepless night he poured into Claude Fable 5.0, Anthropic's model, the full account of what had happened along with screenshots from his phone, and asked whether his symptoms —rumination, self-blame, nightmares— matched post-traumatic stress disorder (PTSD). As he himself recounts in The Free Press, the chatbot repeated the usual disclaimer that it cannot diagnose, but "implied" that his self-assessment was correct, and when he asked what treatment to seek, it suggested cognitive processing therapy (CPT), a clinically backed protocol specific to PTSD.

He explains bluntly why he turned to AI instead of a professional: quality therapy is expensive and time-consuming, and he kept putting it off. It is that cost barrier, rather than enthusiasm for the technology, that drives the experiment. And it comes at a particular moment: Nucci himself notes that US health officials are piloting the use of AI to deliver psychotherapy, proof that the question is no longer hypothetical for the health system.

The most revealing detail in the article is not clinical, it is professional behavior. Nucci describes the experience as "profound" and says it was "truly healing," but makes clear he would not prescribe it to his patients. That contrast —good enough for me, not something I would prescribe to the people I treat— is the line that carries the most weight in the entire piece. A clinically trained therapist knows how to tell manageable rumination from a crisis with a risk of self-harm, knows when to escalate a case and what a patient needs that no language model offers today: continuous follow-up, legal responsibility and a contingency plan if something goes wrong. Apply it to yourself, yes; prescribe it to others, no: that is probably the most honest summary that can be given today of the real state of AI therapy.

Broadly speaking, the use of chatbots as a substitute for or complement to professional therapy has spread rapidly in recent years, almost always ahead of any regulatory framework or rigorous clinical validation. Nucci's account does not prove the pattern is safe to generalize —it is a single case, told by its own protagonist, with no external clinical follow-up— but it neatly illustrates the underlying tension: demand for psychological support is growing, the supply of qualified therapists is not keeping pace, and the cost of a session remains the barrier most often cited by those who need help and do not ask for it.

Our take: the two sides of the thesis we defend at Zendoric coexist here. In the short term the risk is real and far from minor: unsupervised diagnoses, chatbots with no capacity to manage a genuine crisis, vulnerable patients who may mistake algorithmic validation for clinical judgment. But the underlying impulse —the same one that leads a professional who warns against self-diagnosis to do exactly that in the middle of a sleepless night— is the one that in the long run could widen access to mental health care for millions of people who lack it today: cheaper, available at any hour, with no waiting list. The challenge for regulators and developers is not to choose between banning therapeutic AI and letting it circulate unchecked, but to build the intermediate step that is missing today: protocols for referral to a human when the risk demands it, clinical validation of what the model suggests, and clear limits on what a chatbot can and cannot treat. That step, more than the model itself, will decide whether this technology ends up being a safety net for an overwhelmed mental health system or a patch that hides its collapse.