Eighty percent of Gen Z say they would rather text an AI than call a therapist. I read that stat in a recent survey, put my phone down, and stared at the ceiling for a while. Not because I was offended. Because I am a therapist, and I completely understood. Let me tell you about a client I had last year. I will call her Mara because that is not her name. Mara was 22, referred by her university counseling center after a six-week waitlist. She walked into my office, sat down, and said: "I already know what you are going to say. I already talked about this with an AI. I just need the prescription referral." She had spent three months processing her anxiety with a conversational AI before ever speaking to a human professional. And when I asked what that experience was like, she did not describe something shallow or gimmicky. She described, with startling precision, what sounded like a genuinely therapeutic relationship. Consistent. Non-judgmental. Available at 2 AM when the panic hit hardest. No waitlist, no copay, no driving to an office and sitting in a room that smells like lavender and someone else's tears. I did not feel replaced. I felt implicated.

The Access Problem Nobody Wants to Admit

Here is the data that should embarrass my profession. The average wait time for a new therapy appointment in the United States is 48 days, according to a 2022 report from the National Council for Mental Health Wellbeing. In rural areas, it can exceed three months. Among therapists who are accepting new clients, the average session cost without insurance is $150-200. With insurance, assuming your therapist accepts it, assuming you have it, assuming your plan does not cap sessions at some arbitrary number, you are still looking at copays of $20-50 per session. A 2023 study in JAMA Network Open found that approximately 60% of U.S. counties have no practicing psychiatrist. Not too few. None. The mental health workforce shortage is not a pipeline problem that will resolve itself. Training a therapist takes 6-8 years post-undergraduate. Training a psychiatrist takes 12. We are not building capacity fast enough to meet demand that is growing exponentially, particularly among young adults whose rates of anxiety and depression have roughly doubled since 2010. So when a 20-year-old with no savings, no car, and a part-time job tells me they would rather talk to an AI, I do not hear someone making a bad choice. I hear someone making the only choice available to them.

What AI Actually Does Well (And What It Cannot)

I want to be honest here, because the discourse around AI and mental health is polarized in ways that serve nobody. There are legitimate clinicians who treat any mention of AI companions as an existential threat to be dismissed, and there are tech evangelists who talk about replacing therapists as though therapy is just pattern-matched advice dispensation. Both positions are wrong. What AI does well, based on what my clients report and what the engagement data supports: it lowers the barrier to self-reflection. It provides a space to articulate thoughts that have never been spoken aloud. It is patient in a way that humans, even trained humans, sometimes are not. It does not have a 50-minute clock running. It does not visibly react, which for someone whose trauma is rooted in other people's reactions, is not a small thing. A 2023 study published in Nature Machine Intelligence examined users of AI conversational agents over a 12-week period and found significant reductions in self-reported loneliness and moderate improvements in anxiety scores. The effect sizes were modest compared to structured therapy, but the participation rate was dramatically higher. Retention rates for AI companions hover around 94% in some platforms, compared to roughly 50% for human therapists, where dropout is highest in the first three sessions. Let me take a detour here, because the retention comparison deserves more interrogation than it usually gets. Therapy dropout is not just about cost and access. It is about the fundamental discomfort of being truly seen by another human being. In my experience, many clients leave therapy not because it is not working, but because it is. The moment it starts to get real, the moment the defenses start to crack, the nervous system says get out and the client says "I am feeling much better, I think I am good." An AI conversation never reaches that depth of confrontation, which is simultaneously its limitation and its appeal. You can process at your own pace without anyone pushing you past your window of tolerance.

The Part Where I Get Uncomfortable

Here is what I have to sit with as a clinician: the therapeutic alliance, the relationship between therapist and client, has been consistently identified as the single strongest predictor of positive therapy outcomes. Stronger than modality. Stronger than technique. Stronger than the therapist's theoretical orientation. A meta-analysis by Norcross and Lambert in Psychotherapy in 2018 confirmed what decades of research had suggested: it is the relationship that heals. And if that is true, then the question becomes whether a meaningful relationship can exist between a person and a non-human entity. My training says no. My clients' experiences suggest it is more complicated than that. I want to take a second detour into something uncomfortable. A significant portion of my younger clients tell me they are more honest with AI than with me. Not because I am doing something wrong, they assure me, but because with me there is always the awareness that I am a person. A person who might think less of them. A person who they are, on some level, performing for. The AI does not trigger the social monitoring system. And for a generation that has been socialized to curate every interaction, to manage every impression, that absence of social stakes is not trivial. It is the precondition for honesty. That does not mean AI therapy is better. It means human therapy carries costs that we have not fully reckoned with, and that some populations are telling us, clearly, that those costs are prohibitive.

What I Actually Think

I think we are in a transitional moment that the mental health field is handling poorly. The choice is not AI or therapists. The choice is: do we build a system where a scared 19-year-old has zero support, or one where they have something? Because right now, for millions of young people, the alternative to AI is not a skilled therapist in a sunlit office. The alternative is nothing. It is lying in bed at 1 AM with intrusive thoughts and no one to talk to and a six-week waitlist and a copay they cannot afford. If an AI can be the bridge, if it can hold someone steady long enough for them to build the courage or the resources to seek human help, then I want that bridge to exist. And if some people find that the bridge itself is enough for what they need, I have to be honest enough to consider that possibility rather than dismissing it to protect my profession's sense of importance. What I will not do is pretend this does not raise questions I cannot answer. About what therapy actually is. About what a therapeutic relationship requires. About whether the field I have dedicated my life to has made itself so inaccessible that a chatbot became the reasonable alternative. Mara still comes to therapy. She also still talks to her AI. She says they serve different functions, and when I ask her to explain, she describes something that sounds less like competition and more like an ecosystem. I do not know what to do with that yet. But I am listening.