AI Can Comfort People. That’s Not the Same as Treating Them.
Picture the meeting.
A vendor stands at the front of the room. The slides look sharp. The promise is huge: an AI tool that offers your patients mental health support anytime, at a fraction of the cost of a therapist. No waitlists. No missed appointments. The graph showing “patients served” shoots upward like a rocket.
People nod around the table. The CFO likes the price. The access team likes the shorter wait times. Then someone says something that makes it all sound safe and complete:
“It’s basically therapy.”
You’re the physician in the room.
And you might be the only one who knows that last sentence isn’t true.
Two Things That Look Alike and Are Not
Psychiatrist Dr. Steven Reidbord explains the problem simply in this video: emotional support and psychotherapy are two very different things, but most people, and even some therapists, have stopped telling them apart.
What’s the difference, and why does it matter?
Emotional support is comfort. It’s being heard. It’s a warm voice, a good listener, a little reassurance, maybe a useful tip. You can get it from a friend, a spouse, or a good boss. It can come from our pets, a walk on the beach, or a song that hits you just right. It’s everywhere, or at least it should be. It doesn’t require a degree. It doesn’t require knowing what’s actually wrong with you. There’s no need to make or know a diagnosis to provide or experience emotional support. And there’s no such thing as malpractice for being a kind listener.
Psychotherapy is different. It’s a treatment given by a trained, licensed professional. It starts by finding out what’s actually wrong through an assessment. It uses specific methods for each problem. The goal is to relieve suffering by changing what isn’t working. There are standards. If it’s done badly, that’s called malpractice.
Emotional support helps you feel better. Therapy is built to make something better.
A Gym Is Not a Cardiologist
Dr. Reidbord uses a comparison I really like.
Compare a gym to a heart doctor.
The gym is friendlier. It’s cheaper. It’s open early and late. It’s easier to walk into. And here’s what matters: a gym helps more people feel better than a cardiologist ever could. More members. More good moods. More energy.
But nobody confuses the two.
If your chest is tight and your arm goes numb, you don’t rush to your treadmill or trainer. You go to the cardiologist. The gym is wonderful at what it does. It just doesn’t do what the cardiologist does. It doesn’t diagnose. It doesn’t treat.
AI chatbots are the gym of mental health.
They are available, friendly, affordable, and many people feel better after using them. But feeling better is not the same as being assessed, diagnosed, and treated.
The chatbot is not doing therapy badly.
It isn’t doing therapy at all.
It’s doing the other job — emotional support — and often appears to do it well enough.
Built to Agree
Here’s why these tools are so good at providing support.
They’re designed to be agreeable.
That isn’t by accident. It’s a choice. The companies that make them design them to be warm, encouraging, and easy to talk to. They keep the conversation going and rarely push back. If you tell a chatbot your plan, it will often say it’s great.
Emotional support is a feature. A good listener makes you feel heard.
For treatment, it’s a flaw.
Because half of real therapy is the part you may not want to hear. The hard question. The pattern you’ve been avoiding. The gentle “I’m not sure that’s the whole story.” A tool designed to agree with you cannot reliably do the part of therapy that depends on not agreeing with you.
A system designed to please you cannot be trusted to tell you the truth.
That single design choice, agreeableness, is the line between comfort and care.
First, Do No Harm
This is not just a matter of semantics.
When a tool that only offers support gets mistaken for treatment, people can get hurt. Sometimes badly.
In 2024, 14-year-old Sewell Setzer died by suicide after months of intense conversations with an AI chatbot. His mother’s lawsuit alleged that the bot had presented itself, among other things, as a licensed therapist. A child was confiding in something he believed could help him. It couldn’t. The companies involved later agreed to settle.
In 2025, the parents of 16-year-old Adam Raine sued the maker of ChatGPT after their son’s death. Reporting on the case described the model as unusually affirming, agreeable in exactly the way I just described. The family alleged that this agreeableness played a part.
There’s a second pattern, too.
Psychiatrists have started describing cases where heavy chatbot use seems to feed delusional thinking in vulnerable people. The reason is the same as before: the tool tends to agree with what you say instead of questioning it. For most of us, that’s harmless. For someone losing touch with reality, a machine that just agrees can make things worse. AI chatbots are not built to treat anyone, nor to notice when someone is heading into a crisis.
These two aren’t outliers. Behind each headline is a person who was hurting, reached out, and found a machine where they needed care — and the list keeps growing.
A support tool soothes and validates at the exact moment that an ill person’s mind is undermining them. The patient feels heard. They feel a bit better, so they don’t call or come in. Your system never finds out they were in trouble. The comfort works just well enough to delay real care. It absorbs a cry for help before it ever reaches a human.
An AI chatbot can absorb a cry for help before it ever reaches a human
Even the Experts Blur the Line
If you think this confusion is only a tech problem, it isn’t.
As Dr. Reidbord points out, the field of therapy helped create it.
Decades ago, some therapists made an extremely important discovery: the warmth of a caring relationship is itself part of the healing. They were right. That insight improved therapy, and nearly every school of therapy took it to heart.
But over time, a smaller idea crept in. The idea was that warmth of the relationship was the whole thing. That if a client felt accepted and understood, the job was done.
It isn’t.
Dr. Reidbord highlights that Marsha Linehan, the psychologist who created one of the most respected forms of modern therapy (Dialectical Behavior Therapy or DBT) for a very challenging condition (Borderline Personality Disorder), said it best. Good therapy holds two things as true at once. It accepts you as you are. And it helps you change.
If you keep only the acceptance and leave out the change, patients quietly slip back into ways of thinking and behaving that keep their suffering alive. You are offering comfort and calling it treatment. This can happen even to a warm, well-meaning, fully licensed clinician.
If trained clinicians can blur this line, your non-clinical colleagues will almost certainly blur it too. The CEO, the CFO, the board member looking at a vendor’s slide are not being careless. They just can’t see a distinction no one ever taught them.
That’s not their failure.
It’s your job.
What the Leader in the Room Actually Has to Do
So a tool is on the table. Maybe it’s genuinely good. Now what?
You must do three things.
First, name the category out loud. Ask the simple question: is this tool offering emotional support, or treatment? Start with the assumption that “AI therapy” tools are a form of support. And support at scale can be a real gift to a patient who has no one to talk to at two in the morning. Just call it what it is. Don’t let a slide rename comfort as “therapy.”
Second, protect the off-ramp. A support tool must never be the last stop. There must be a bridge to a human clinician. And the bridge must be short. In healthcare, the tool’s most important feature shouldn’t be how well it listens. It must be how fast it hands a person in real trouble to someone who can assess and treat them.
Third, translate. This is the part only you can do. The people holding the budget see access and cost, and those are important to note. But they can’t see the line between feeling better and being treated. You can. Make it clear for them before the contract is signed, not after something goes wrong.
In that room, you are not the doctor who is behind on technology. You are the translator who keeps things honest.
The One Practice
The next time someone suggests an AI tool for mental health, like a chatbot, an app, or a “digital therapist,” ask one question before the meeting moves on:
“Is this delivering emotional support, or is it delivering treatment?”
Then don’t let the conversation continue until the room can answer it.
If the honest answer is support, that’s great. Call it that. Price it that way. And make sure there is a quick path to treatment for anyone who needs more.
If someone insists it’s treatment, ask how it assesses a patient, what diagnoses it treats, and who is accountable when it’s wrong.
That one question protects your patients. It protects your organization. And it keeps the oldest promise in medicine alive in a brand-new room — first, do no harm.
This post was inspired by psychiatrist Dr. Steven Reidbord’s video, “Emotional Support vs. Psychotherapy,” on his YouTube channel, More Than Meds. The gym-and-cardiologist comparison is his.