Will AI Replace Therapists? What the Research Says.

I see the subtle scrunch between her eyebrows. I notice that slight stiffening of his shoulders as his wife describes their weekend. I hear the catch where her emotion got stuck in her throat before the sentence finished.

And I know that this is the sentence to pause on. There’s something underneath it. Forget bullet points in the treatment plan or the next worksheet in the protocol. The thing that just happened in the body of the person across from me, is the work.

It’s also the part I’m not sure AI will ever be able to truly do. It may read an algorithm that calibrates the changes in posture, but will it ever be able to truly get, on a visceral level, what that subtle micro-movement feels like to a person who is bearing their most vulnerable self to another person?

I have been thinking a lot about AI in therapy, the way most of my colleagues are. The reflexive position in our field right now seems to be defensive. AI is coming for us. We have to fight it. We have to prove that what we do is real and irreplaceable, lest the bots take our jobs and our credibility along with them.

I understand the impulse, but the more I think about it, I don’t think it serves us.

What I want to do instead is something closer to what I would do with a client grappling with a hard moment. Sit with what is actually happening. Notice the reaction. Where I feel it in my body. Find the parts of it that are useful and the parts that are not. And then decide what to do from there.

Two clinical tools in particular keep coming to mind when I think about AI in therapy.

The first is the principle that two things can be true at once. In my world we call that a dialectic. AI has real and concerning implications for our society, our privacy, our mental health, and the environment.

And what AI can do is genuinely remarkable. Both are true and they’re not mutually exclusive.

The second is radical acceptance. AI in mental health care is an inevitability (it’s already here). It is not a question of whether, only of how. Radical acceptance does not equal complacency. But it frees up the distress we experience fighting what is, and focuses that energy on what we can do.

“Grant me the serenity to accept the things I cannot change. The courage to change the things I can. And the wisdom to know the difference.”

Once you sit with both of those concepts, the question changes. It stops being "how do we keep AI out of therapy" and becomes, "what is therapy actually for, and what part of it is uniquely ours?"

What AI Can (and Should?) Do

AI is going to win the worksheet wars.

It can produce a CBT thought record in seconds. It can draft a behavioral activation plan, a values clarification exercise, a graded exposure hierarchy faster than it’ll take you to finish reading this post. It can pull a relevant study faster than any clinician. It can summarize a treatment manual in the time it takes you to make your coffee.

This is not speculation or a future concern. This is happening in real time, and the technology advances daily. 

Setting aside legitimate concerns about what AI will do on a macro level across all domains of our existence, if we are being honest, it’s a very useful existential threat. AI as a clinical assistant, with appropriate human oversight, could give therapists back enormous amounts of time. The hours spent on documentation, session prep, hunting down the right worksheet, writing psychoeducation content. That is time we could spend going back into the work. Or back into the therapist's own life.

Human therapists are not going to outpace AI on the procedural elements of the job. We should stop trying.

What is Actually Ours 

So what do we have that AI doesn’t? The part of the work that the field has spent recent years increasingly devaluing. The art of therapy.

Our intuition, attunement, and the capacity to truly empathize with another person, because we’ve been there. Our clinical instincts for when to affirm and when to challenge, and our willingness to sit in a hard moment without rushing to fix it. The ability to laugh at a dark joke and notice the new haircut.

Call them soft skills if you must, but the art of therapy is the skill that matters most. While we’ve been chasing validity as a science, the research has been consistent.

Decades of psychotherapy outcome research and literature invariably find that the single strongest predictor of success in therapy is the quality of the therapeutic relationship. The bond and rapport between a clinician and their client. Whether they are a good fit. Whether the person feels seen, safe, and met. This holds across modalities, diagnoses, and orientations.

The technique matters. Science is amazing. But it simply matters less than the art.

How We Got Here

We didn’t arrive at the worksheet-heavy, protocol-driven version of therapy by accident. We got here for reasons that made sense at the time, even when the result has been costly.

The soft sciences had to get hard in order to be taken seriously. To compete for funding, for credibility, for parity with medicine, psychology had to demonstrate that what we do is measurable, replicable, and evidence-based. The push toward manualized treatment was a real and good-faith effort to legitimize the field. Treatment manuals make studies possible and studies make a profession look like a science.

Then insurance companies arrived with their own demands. To pay for therapy, insurers needed medical necessity justified by symptom checklists, treatment plans, and measurable outcomes. It needed to be time-limited, code-authorized, and documented to perfection to be sanctioned for reimbursement. The financial pressure on the entire profession to look insurable pushed therapy further into the procedural shape it has today.

It is what happens when a profession needs to survive inside a market that only knows how to value what can be checked in a box.

The result, though, is that we trained a generation of therapists to believe that the formula is the work and the manual is the medicine. That if you follow the protocol with fidelity, the outcome will follow.

It will, sometimes. The science is real. The evidence base matters. I am not arguing against any of that.

I am arguing that the science was always supposed to be the floor, not the ceiling. And we have been working on the floor for a long time.

Both, And

Two things are true. The science of therapy matters and the art of therapy is what makes it medicine. The arrival of AI may be the most useful pressure the field has had in decades. It may actually force a conversation about what therapy is actually for, and what part of the work cannot be replicated by a model.

If AI takes over the worksheets, protocols, and procedural elements, what is left is the relationship. The attunement. The art.

Which, again, is the part the research has been pointing at all along.

So while I understand the concerns in the field right now, I think we have it backward. AI is not the enemy of good therapy. AI is the enemy of the parts of therapy that were always more about reimbursement and legitimacy than about healing. And maybe, if we let it, AI will do us the favor of clearing those out so the actual work can come back to the center.

That will require us to take the relational work more seriously again and to stop treating it as the warm-up before the real intervention starts. To start treating it as the intervention itself. And based on some of the trends I’m seeing, many clinicians are hungry for this realignment. 

But, it will require the people who fund and regulate mental health care, the insurance companies, licensing boards, and credentialing bodies, to recognize that what makes therapy work is not always what fits cleanly on a form. That will be a longer (harder) conversation. But it’s the conversation we are going to have to have.

This is the first in a series I am writing on the art of therapy in the age of AI culminating in a bigger frame this all sits inside. More to come.