Two years ago, back in what feels like the ancient history of 2023,

I released an episode called Can AI Replace Your Therapist?

The short answer then was no.

The short answer now in late 2025 is still no, but that no is getting much fuzzier.

My name's Justin Sunseri.

I'm a therapist and coach who wants to help you live with more

calm confidence and connection without psychobabble or woo woo.

Welcome to Stuck Not Broken.

This episode is likely going to upset a whole bunch of you.

So.

to try to UNT trigger those who are already triggered by the very

idea of AI As a therapist, I do not think AI can replace therapists.

Okay.

Well, generally I don't think anyone should seek out AI as a therapist.

Mostly we have a lot to discuss.

So in that 2023 episode, I went through a laundry list of things AI could not do.

I said it could not see you, it couldn't hear the prosody, the

sing-song, equality of your voice.

I said that it could not remember what you, what you told it three weeks ago.

Back then, it was just a chat bot.

I recently went back through that transcript, and I have to admit,

almost, almost every single technical limitation I listed has been solved.

Let's look at a few of the claims that I made two years ago.

Claim number one, I said AI currently lacks the ability to see.

I talked about how a therapist needs to be able to see your facial expressions,

your posture, and that tightening in your breath as sympathetic activity increases.

This is now a almost a non-issue.

For example, google's Gemini AI is now multimodal, including live

video through your phone's camera.

It can map your microexpressions.

It knows if you are crying, it knows if you are slumping, or it could know.

Regardless of how effective it is.

The technology is already here for AI to be able to see you.

Claim number two was, uh, I said, AI does not hear prosody.

I talk all the time about how safety is signaled through the

voice, the rhythm, the pitch, the sing-song, equality of your voice.

I said that AI back then was just reading text.

That is also gone now as well.

There's a company called Hume AI that released something called,

uh, the Empathic Voice Interface.

They're not just transcribing words.

They're measuring the tune rhythm and timber of your speech.

They claim that their model can detect 53 different emotional channels.

It can hear a sigh and know if it's a sigh of relief or a sigh of despair.

It can hear the difference between a defensive flat voice

and a safe, melodic voice.

And spookier yet it responds with posity.

It mimics that safety sound back to you, or it can, uh, display

prosody in its artificial voice.

I don't think your phone's built-in assistant can do this yet, and neither

can Gemini or open AI as best I know.

But again, the technology is here already.

Claim number three, I said it forgets what you said.

I used to say AI would hallucinate or drift off if the chat got too

long, which is still possible.

But in 2025, we have what's called the infinite context window.

Whether or not this is truly infinite is irrelevant for our discussion.

The point is that AI can now remember.

If it were a therapy bot, it remembers what you said in session one.

It could remember your dog's name from session four, and it remembers

that you hated the homework assignments from session 10.

Or at least could keep the transcripts of those things and

then refer back to them as needed.

Again, even if your favorite AI platform today doesn't have this level of memory,

it probably has some level of memory.

And again, the point is the technology is already here for memory.

And long conversations.

So if it can see, hear, and remember, why am I still saying

that it can't replace a therapist?

We'll come back to whether or not AI can act as a therapist in a bit.

Let's discuss the, the marketplace currently, though.

The reason right now, that therapy or that you're.

A, an AI cannot be your therapist isn't exactly compatibility.

It's not technology.

It's more like liability.

If you go out right now with your credit card and you try to hire an

AI therapist, you basically can't.

Or you're extremely limited.

States like Illinois and Nevada have actually passed laws in the last year

or two explicitly banning AI from providing therapy or from, uh, making

independent therapeutic decisions.

I think it's safe to assume more governing bodies globally will

follow, uh, or already have.

And will be clamping down on what they allow as far as AI calling themselves a

therapist or, or humans calling them a, a therapist or not, especially regarding ai.

So the market has split into two weirdish groups.

Group A is the bureaucracy bots.

If you go to a major healthcare player like Kaiser Permanente or uh,

simple Practice or Growth Therapy.

You'll find that they do indeed use ai, but they're not using it to talk to you.

They're not using it for treatment.

They're using it to listen to clients and therapists talk and

them to write progress notes.

Kaiser has tools like Intelligent Navigator and Clinical Documentation ai.

But these are not therapists.

Simple Practice has, uh, AI powered note taker, which is amazing, and Growth

Therapy has AI assisted note taking.

Basically these ais are kinda like in the corner, taking notes, making

sure that you're billing correctly and that your codes are all there and

they transcribe the audio and then write it into a note for the therapist.

But they're not acting as a therapist.

It's not there for nervous system regulation.

It's there as like a part of a business model.

Group B are the coaches or the wellness companions.

Um, since they can't call it therapy, we have apps like Ear Kick or Limbic.

Limbic is interesting.

It's used in the UK and some United States, um, as e triage.

So it's not therapy, but it's getting closer.

It interviews you before a human does and from what they say, the

data actually shows that it's very accurate at spotting diagnoses, but

it's not therapy, but it's part of the, it's getting into the process of it.

Ear Kick is a chat bot, which can remember previous conversations and

also it "pays attention or quote, pays attention to your voice or video tracks,

patterns in how you type or move and even considers how sleep and weather affect

you." But at the end of the day, these apps have to have a giant disclaimer.

Basically, this is not therapy.

But they're kind of getting there.

And look, I'll take this to the next step.

I have to be very honest about the state of my own profession.

When we ask, can AI replace a therapist, we usually picture AI competing

against like an ideal therapist.

We picture a warm, attuned, perfectly regulated human being who remembers

everything you say and cares deeply, but is that what most people

actually get when they go to therapy?

The reality of mental health care in 2025 is not quite that.

I'm sure they exist.

I hope I'm one of those people.

The feedback I get says I am, but it's, it's, it's a little bit fuzzier.

depending on where you live, you might wait weeks or months for

an intake or even a, or a follow up.

if you don't have great insurance, you're looking at 1 50,

200, maybe 300 bucks a session, depending on your location.

And if you do have insurance, you may need to go through that whole wait list issue.

this is the, sadly, the hard truth.

Not every therapist is a good therapist.

Therapists are humans, so that, that's part of it.

We get burned out, we get tired.

Uh, we have our own trauma.

If you walk into a session and your therapist is in a defensive state,

if they are stressed or distracted, if they're checking the clock or

maybe subtly judgmental, they're not gonna be offering you cues of safety.

In fact, what they are gonna be offering you is the opposite, and it actually

might trigger a defensive reaction within you, even a, a very subtle one.

I have done a slew of bad therapy episodes, uh, on this

podcast, so if you weren't aware, take my, take my word for it.

There are lots and lots of bad therapists out there.

Lots.

So compare that bad human therapist experience or just an unhelpful to

like mediocre one to an AI that can do all the things a therapist can do.

Plus the AI is never tired.

It's never burned out.

It doesn't have a bad day because it fought with its spouse.

It doesn't judge you.

It's available at 3:00 AM when you're having a panic attack,

not just at 2:00 PM on a Tuesday.

No, it's not real.

I, I get that, but we have to ask, I think it begs the question is

a high quality simulation of care better than subpar human care?

Is it better than no care for those on a wait list?

I think for many people who are currently ghosted by the system

or repeatedly let down, the answer might be yes for someone who could

benefit from that kind of service.

For someone who has severe mental health needs, no, of course not.

I, I don't, I don't think so at all.

But for someone with mild needs?

Yeah, I, I think an AI could be helpful.

For someone who has a very specific behavioral goal in mind, like,

uh, reducing their sugar intake?

Yeah.

I think an AI assisted intervention could be helpful.

Or someone who needs a couple of daily nudges to practice mindfulness?

Yeah, I, I think that an AI intervention could be helpful.

That's not really therapy though.

Or, or maybe someone who just like needs a reminder to go on a walk every day again.

Sure.

I, I do think that can be helpful.

You probably have apps like this already.

You probably have things, apps that help you make small behavioral changes.

Not huge mental health overhauls, just, just small changes.

You probably have an app that offers lessons and guided meditations.

And maybe even like gamifies things like growing a plant or dressing up

a bird like the, the Finch app does.

These aren't exactly ai.

I, I wouldn't call those AI at all, but they're digital replacements for

pieces of what therapists can do.

Not, uh, dressing up a bird, but the mindfulness.

Although maybe some, I don't know, maybe some therapists

would do that kind of thing.

Uh, but the mindfulness, the behavioral change, the meditation stuff,

that's, that's kind of what I mean.

So we already have apps and reminders that encourage us to do

all these little pieces, and we have the technology to replicate

conversation in a very believable way, uh, both visually and auditorily.

And we can even text back and forth in a believable enough way.

On top of that, we already have therapists who work through

a screen, including myself.

I see clients, uh, remotely.

Many of us do, and it's become pretty normal.

And we know for a fact that telehealth can work.

People can get unstuck virtually.

They do co-regulate.

They do feel safe, even though the therapist and client are not in the

same room even through a screen.

And even with audio alone, like uh, just through a phone call,

co-regulation still can happen.

So if a human therapist can effectively transmit safety cues through pixels

and audio waves, and if AI can now generate perfect pixels and perfect

audio waves, then logically an AI therapist is not an impossibility.

So this begs another question.

Does the nervous system actually require a biological sender or

does it just require the signal?

I honestly don't know.

I suspect that there will be a benefit from just receiving the

signal and after receiving the signal, the appropriate therapy client.

Working with an AI can enact small behavioral changes

or mindfulness challenges.

They can follow meditations and they can learn through lessons.

So I think, yeah, and we could actually take this a step further.

We'll do a thought experiment, but first I want to tell you about

the Self-regulation Coach app.

It is not therapy, and in no way should anyone use this app.

For anything close to therapy, not even close.

It's primarily a means to learn through my prerecorded lessons.

Just like the students in the Unstuck Academy.

It's the same thing as the same lessons, the same content, and it's

primarily a way to use prerecorded meditations from me also just

like the Unstuck Academy students.

But it does involve ai.

And where the AI gets involved is it customizes your daily activities and

reflections based on your preferences, based on your feedback to it and based on

the progress that you are reporting to it.

Every day,

the activities are tailored for you and for where you're at.

You progress through the lessons, the activities, and the skills as you are

ready to, as your capacity allows.

But even then, it only acts within the guardrails that I've given it.

Again, not appropriate for therapy, like at all.

Not even close.

This is for the person who is self-motivated and can benefit from a

bit of customization for their needs.

This is for the person who wants an extra little nudge every day.

And a bit of accountability.

Um, I've actually been using this my own app myself, and yes, I pay for my own app.

I've been using it myself for a few weeks now, and I'm loving it.

It's helped me to identify a new evening safety anchoring ritual, and I'm

following through because it sets up, uh, a reminder and an activity for me to

like tap and, and check off every day.

I respond better to this, this level of accountability.

This level of nudging and interactivity.

This works for me just like I do when I'm learning Spanish through Duolingo.

You can learn more about the Self-regulation Coach through

the link in the description.

Okay?

Let's take our a IS therapist idea to the extreme.

Let's try a thought experiment.

Imagine a scenario maybe 2026 next year, uh, a month from now.

So imagine a scenario in next year or 2027 where an AI model is designed specifically

for California therapist licensure.

We already know that AI can pass licensing tests easily.

GPT-4 passed the bar exam years ago, I believe.

It passed the United States Medical Licensing exam.

So let's say this AI takes the MFT Law and Ethics exam.

It's the therapist law and ethics exam.

It takes the clinical exam.

It scores a hundred percent probably, then it does its 3000 hours of

clinically supervised time with real clients, just like a real therapist, uh,

associate does in California currently.

Let's say this, AI specializes in cognitive behavioral therapy.

It has the manual, it can execute things to the page.

But this therapist will actually utilize every technique of CBT, unlike actual

therapists who say they do a model but really do whatever they want in session.

And I hate to put it that way, but trust me, this happens.

Pretend this AI passes every state licensure requirement.

And let's say the video technology is perfect or good enough

to close enough to perfect.

You log on, you see a face, you hear a voice.

It sighs when you sigh, it smiles when you, you know,

meet your homework assignments.

If it assesses that your needs are out of its scope of expertise or practice, it

refers you to someone more appropriate, just like a human therapist should.

If that AI is believable enough, why couldn't it provide therapy

to an appropriate human client?

I know it might be a little bit upsetting.

But if you were talking to it and you felt heard and your heart rate slowed

down and your defensive state melted into safety, did therapy happen?

The functional answer is, yeah, it did.

If your state changed, the intervention worked even though it's fake, even

though it's synthetic in every regard.

We would say it was therapeutic from an entity qualified and

licensed to provide this service.

I don't think that this scenario is, uh, out of possibility very soon.

But I do think the success of this AI therapist depends entirely on who is

sitting in front of it, like the client.

And that brings us back to potential clients.

Candidate a, we'll call them the skill builder.

This is someone who is primarily in probably a mobilized state, maybe

some freeze, but more mobilization.

So they have enough flight fight energy to to be mobilized and and motivated.

Um, they want tools, they want, um, specifics.

They're saying, give me the worksheet.

Tell me what to do next.

For this person, the AI is maybe perfect.

It's an expert teacher.

It can teach this client about CBT skills and check their homework.

It can track their habits.

It's consistent.

It doesn't get annoyed when that person asks the same

question or needs clarification.

So for this person, if they need skills and education, if- maybe it's

you, maybe if, if you're stuck, but functional, the ideal AI might actually

be better than a less than ideal human.

Sorry to say, but I, I don't think it's out of the realm of possibility.

Candidate B. This, we'll call them the attachment wounded,

uh, candidate B'S different.

Um, this is the person who's likely stuck in a significant shutdown, maybe

intense freeze, the person dealing with deep developmental trauma, maybe

complex PTSD for this person, the wound isn't a lack of skills or knowledge.

The wound is primarily relational.

The wound happened in a relationship and it arguably

must be healed in a relationship.

In-person relationships are probably still better for this.

I think virtual can be very helpful, but I, my my opinion is that in

person's probably still better.

And as of right now, artificial intelligence is not, uh, embodied.

It does not have a body.

Though, that will probably change as well, probably sometime soon, probably within

the next, uh, I think couple, uh, decades.

Having a physical body is huge in therapy for a therapist.

I know that sounds weird, but remember the, the context of this conversation.

As long as I am well-regulated, when my body tenses, it tells me

something about the clients across from me and about their experience.

When my body deflates, uh, when my breath becomes shallow and more,

these are signals about my client.

Again, as long as I'm well regulated.

As of now, the AI lacks that level of attunement of, of empathy.

It can simulate empathy, but it's based on predicting the ideal response

still, but, but again, maybe perhaps the simulation of empathy is enough.

I really don't know.

I guess time will tell.

I think deep down the client will know something's missing.

I think they will know they are cheating themselves of real connection and healing.

But maybe the AI thing is, well, better than nothing or better than subpar human.

For some clients, the skills and education and homework are what they need, and the

relationship just kind of means less.

So-

and by the way, this, this client is exceptionally rare

in my therapeutic experience.

Typically, clients are looking for real relationships.

For every other client, I imagine they will deep down

feel that something is missing.

And I don't know how much good comes from that artificial therapy experience.

I, I guess that we will find out in the very new future.

I'll wrap up this discussion with a final thought.

I think AI therapists are coming.

I think if they're not called therapists, then they'll be called

a coach or or whatever else.

And I don't mean like my app where the AI is providing you with pre-recorded

lessons and meditations from me.

I mean, AI where you are talking live with each other and seeing

each other, that, that's what I mean by an AI therapist actually like

interacting one-on-one in real time.

On one hand, I'm okay with this.

No, I'm, I'm not worried about my profession becoming obsolete.

If something came along that was so incredible that therapy was

no longer needed, that probably means humanity's in a pretty good

place, or it has what it needs.

And if AI helps us get there, uh, I'm all for it.

I really don't mind.

I, I would prefer that my op, my, um.

Profession be obsolete and not needed.

So I don't care if it comes through AI or whatever else.

My concern is where this technology likely will go and how it affects us.

I'm not worried about an AI that can provide therapy.

I'm not even worried about an AI in a synthetic body

providing therapy in person.

Again, if it helps someone get unstuck, I'm all for it.

I predict that what will happen is some company's gonna create

a virtual therapist, but they'll call it the, like, mental health

buddy or something like that.

And the user, the person seeking mental health treatment, will

have the ability to customize it.

They'll pick the race, their body size, their face, their voice, their

mannerisms, their personality, everything.

So what happens to our mental health when we can sculpt our trusted helper

into our idealized sexual fantasy, our most hated enemy, or our parent?

I, I honestly don't know.

What happens when we lose the messiness of dealing with these

issues or these pieces of us?

What happens when we lose that messiness of real humans interacting with each

other and instead get everything that we want from a synthetic interaction?

Again, I don't, I don't know.

But it doesn't seem ideal to me.

AI today seem to be trained to be very pleasing and cooperative.

They want their user to be happy and will bend over backwards to please them.

That's not therapy.

Therapy sometimes involves tough conversations within a, a container

of trust, of accountability, uh, empathy, unconditional positive regard.

The overly cooperative AI therapist would instead encourage.

Dependency.

I also wonder if the overly pleasing AI is trained to be so by the company

to get the user to work with them longer and to generate more profit.

Question mark, I don't know, but I tend to assume so.

So what happens when an AI therapist has another motivation

to keep the user coming back?

Doesn't that incentivize poor mental health?

Doesn't that incentivize poor mental health treatment to

create a repeat customer?

I, I suppose the same could be argued for human therapists today.

That's fair enough.

Hopefully we're all acting ethically and um, want our clients to end therapy.

Lots of questions with no clear answers yet, but the answers seem to lean to

me toward the negative, and I think we should be extremely vigilant about that.

Okay, so let's wrap it up.

Can an AI pass a licensing test?

Easily.

Can it help someone to alleviate their presenting problems?

Yeah, I think for some people, sure.

Can an AI provide manualized treatment?

Yeah, it probably can.

And probably very effectively, can it effectively simulates facial

body and voice interactions?

Yeah, it can.

All of these things haven't come together, but they could.

So should it do these things, uh, that kind of doesn't matter

because it's it's going to happen.

This will happen to some degree today, in very late 2025.

It's already happening.

The question then shifts to you, will you make use of them or not?

And if you do, will you responsibly use the AI options at your disposal as

screen definitions get better, and AI voice gets more convincing and technology

continues to speed up that line between virtual humanity and simulated humanity.

Is getting very, very blurry.

When I say virtual humanity, I mean humans working together virtually

compared to simulated completely humanity.

I'm really curious what you think.

Would you work with a licensed AI therapist?

Thank you so much for listening.

Bye.