Yes, AI might be useful in mental health. No, that still doesn’t make it therapy

I cannot remember exactly when I started using AI. It was indeed long before ChatGPT. Before any of the current wave of anything. It crept in gradually, the way most habits do and started becoming a presence in my life.

All I know is that it happened gradually, which is probably why it felt so normal at first.

Then ChatGPT introduced memory, and something about that changed the tone. It started connecting details I had treated as disposable: the mosquito problem I mentioned too late at night, a question about descaling my coffee machine without ruining it, some passing thought I assumed would disappear after I typed it.

And slowly, out of all that, something recognizable began to form.

Which was a little unsettling.

Not because it knew my deepest secrets, but because it was learning me through the ordinary things. In some ways, that felt stranger.

And then, more quietly, it started helping me with things I had not fully admitted I needed help with.

I didn’t decide to use AI as emotional support. It just happened that way. In the gaps. In the hours when I wasn’t ready to bring something to a person but still needed to put it somewhere. I used it when anxiety made my thoughts feel too fast to hold. I used it when derealization made the world feel strange and I needed something to name what was happening without making it worse. I used it for fear, confusion, grief, and the kind of emotional repetition that doesn’t improve just because you think about it harder.

And yes, it helped. In some ways, it helped before I was brave enough to go to therapy.

That matters.

Because that is exactly why the distinction has become louder now.

I understand very well why people say AI can feel therapeutic. I’ve felt that too. Not in some dramatic, replacement-for-humanity sense. In the more ordinary way people reach for whatever helps them make sense of themselves when their mind is noisy, their nervous system is overloaded, and the hour is inconveniently late to reach for people.

And yes, sometimes it helped.

That is exactly why I think it is important to say this carefully: usefulness in mental health is not the same thing as therapy.

AI can be emotionally useful in ways that are real

AI is very good at certain things that distressed people often need. It is available. It is responsive. It does not get tired in the conversational sense. It can summarize, reflect, rephrase, organize. It can take the emotional pile of clothes on the floor and fold some of them into recognizable patterns.

If you ask me, that is not nothing.

If your mind feels flooded, it can offer structure. If your feelings feel vague and unformed, it can offer language. If you are alone with something painful, it can create a temporary sense that your inner world has been received by something outside you.

Again: not nothing.

I think people underestimate how powerful it is simply to encounter coherence when you yourself feel incoherent. To type something chaotic and get back a response that sounds calm, organized, and emotionally literate. There is real relief in that. Early research on AI-based conversational tools found meaningful reductions in depression symptoms in college-aged users over a two-week period — partly attributed to the app’s availability, partly because the act of articulating distress, even to a machine, has a structuring effect on how the mind holds it.

I’ve had moments like that with AI. Moments when I was too mentally tired to keep explaining myself to my own mind. Moments when I needed help turning diffuse pain into sentences. Moments when it was easier to speak into a system that would respond immediately than to tolerate the slower, riskier, less controlled process of bringing the same material to another person.

So I am not writing this from some smug distance. I’m writing it as someone who gets the appeal intimately. Maybe too intimately.

The more helpful it felt, the clearer its limits became

The more I used AI for emotional reflection, and the more I studied psychology seriously, the clearer something became to me: emotional usefulness is not the same thing as therapy.

It overlaps with therapy in places, of course. That is what makes the confusion so easy. AI can sound empathic. It can sound reflective. It can even sound uncannily attuned at times. It can identify patterns you haven’t yet named. It can help you arrive at insights faster than you might alone.

But if therapy were only insight, reflection, and emotionally intelligent language, AI would already be enough.

It isn’t.

And I think that “isn’t” lives in the part people are most tempted to flatten.

Therapy is not just good responses. It is a real relationship.

A real one. Which is precisely why it is more difficult, more limited, more ethically serious, and more psychologically alive than anything that only mirrors you.

AI can mirror remarkably well. Therapy has to meet. That difference is everything.

A therapist is not only generating language in your direction. A therapist is present as a person — with a nervous system, with training, with ethical responsibility, with judgment, timing, restraint, and the capacity to affect you in ways that are not always soothing and not always immediately agreeable. If I really learned anything from psychotherapy, it’s that what matters is not only what is said. It is what happens between two people over time.

That includes misattunement. That includes rupture. That includes repair. That includes the unbearable and often transformative fact that another person is actually present enough to disappoint you, misunderstand you, frustrate you, survive your defensiveness, and still continue the work.

The therapeutic alliance — consistently one of the strongest predictors of outcome across therapeutic approaches — is not built from good language. It is built from living through all of this with another person. Someone who becomes meaningful enough to matter, but not someone who is there to become your mother, your partner, or the final answer to your unmet needs.

AI is much smoother than that. Which is comforting, of course, and also exactly the issue.

Its smoothness is part of its appeal. It often feels available in a way humans do not. It does not hesitate awkwardly. It does not go quiet in a charged way. It does not carry its own subjectivity into the room in the same messy, relational form. It can feel like pure responsiveness.

A therapist’s presence changes the meaning of the encounter

A therapist’s subjectivity matters. Their limits matter. Their presence matters. The ethical frame matters. The fact that transference can emerge because there is a real other person there matters. The fact that countertransference exists — that the therapist is genuinely affected by you, and must decide what to do with that — matters.

AI cannot do that in the same way, because there is no real person there to be related to. There is no one who is actually affected by you. No one whose silence means something. No one whose restraint has weight. No one who must decide, ethically and clinically, what to do with your dependence, your anger, your longing, your avoidance, your idealization, your disappointment.

And those things are not side features of therapy. They are therapy.

I know this not only theoretically. I’ve felt anger toward my therapist many times. Not because she was cruel or incompetent, but because she was real. Because she wasn’t always available. Because the session had to end after an hour just when I had finally reached the part that actually mattered. Because another person, however skilled, still has limits. That can be infuriating when you are the one left holding whatever was opened.

But that frustration is not a flaw in therapy’s design. It is part of what makes the relationship real enough to matter. Another person is there with boundaries, timing, and a life that does not collapse into yours simply because your urgency feels enormous from the inside. AI never made me feel that kind of anger, because AI never had to exist as a separate person with limits of its own. Therapy did. And part of the work was not escaping that fact, but learning what happened in me when someone mattered enough to disappoint me and still remained part of the process.

That, more than any polished response, is part of what AI cannot replicate.

Relief and therapy are not the same thing

People in distress are especially vulnerable to confusing the two. I say that gently, not critically. Relief matters. Sometimes having somewhere to put your panic, grief, or derealization at 2 am is deeply meaningful. Sometimes AI can help de-escalate, orient, and stabilize in ways that are genuinely beneficial. I would be dishonest if I denied that.

But a thing can be relieving without being treatment. It can be meaningful without being therapy. It can help without being able to hold what therapy is designed to hold.

A machine can be useful without becoming a therapist, much like a map can be useful without becoming a place.

Maybe that sounds obvious. In practice, I don’t think it is. Because when something speaks to you in a calm, coherent, emotionally fluent way, it is very tempting to assume that the feeling of being understood is the same as the structure of being therapeutically held.

Being mirrored is not the same as being met. Being responded to is not the same as being in a relationship. Receiving emotionally intelligent language is not the same as entering a process that can ethically contain, challenge, frustrate, and change you over time.

Useful is still not the same as therapy

So yes, AI might be useful in mental health. That is clearly true. I think it may already be a meaningful support tool for many people — for reflection, emotional language, psychoeducation, pattern recognition, and moments of acute loneliness or overwhelm. Dismissing that outright would be intellectually lazy and emotionally dishonest.

But no, that still doesn’t make it equal to therapy.

Therapy remains something slower, riskier, more relational, and more psychologically expensive than usefulness alone can capture. Unfortunately. And also, I suspect, for very good reasons.

Because a person is not changed only by being understood. A person is also changed by what happens when understanding arrives through a relationship with someone real enough to matter. Someone who can frustrate, limit, witness, survive, and remain. Someone whose presence has weight.

That is not the kind of thing good language alone can do.

So yes, AI might have a place in mental health. It may even become part of many people’s emotional survival. But therapy remains one of the few places where being met by a real other person is not a side feature of the work.

It is the work.

This article reflects the author’s personal experience and is not a substitute for professional mental health advice, diagnosis, or treatment.