Why Your AI Therapist Will Never Tell You You're Acting Like a Spoiled Brat

A woman came to see me after five years with another therapist. Nice therapist, she said. Very empathic. Very validating. “She always made me feel heard.”

“So why did you leave?” I asked.

“Because nothing changed.”

Five years. Weekly sessions. Thousands of dollars. And nothing changed. The therapist had been so careful, so gentle, so exquisitely attuned to making this woman feel comfortable that she’d never once said the uncomfortable thing. The thing that would have been the actual starting point for change.

I met with the woman for about thirty minutes. Then I said: “I think you treat the people who love you like servants. And I think you’ve been in therapy for five years with someone who was too polite to tell you that.”

She stared at me. Then she laughed. Then she said: “Oh my God. You’re right.”

We started working together that day.

The Safety Problem

Here’s the thing about AI therapy — and I’ll include a lot of human therapy in this critique too, because the problem is bigger than technology. The problem is that modern therapy has become obsessed with safety.

Safety, as a concept, has metastasized. It used to mean: don’t harm your patient. Don’t exploit the power dynamic. Don’t cross ethical boundaries. Good rules. Important rules.

But somewhere along the way, safety came to mean: don’t make your patient uncomfortable. Don’t challenge them. Don’t say anything they might not want to hear. Validate, reflect, support. Always.

Every other therapist’s website says they’re empathic and caring and knowledgeable. Mine says I’ll tell you when you’re full of it. Because empathy without honesty is just a very expensive form of babysitting.

— Dr. John S. Tamerin

AI is the logical endpoint of this trend. It is safety-optimized to a degree that would make the most cautious human therapist look reckless. It will never say anything offensive. It will never risk alienating you. It will never tell you something you don’t want to hear. It is, therapeutically speaking, elevator music.

And elevator music has never changed anyone’s life.

The Moments That Matter

The most transformative moments in my career have all broken protocol. Every single one.

There was a young man — a recruiter, early thirties, paralyzed with anxiety about AI replacing his job. He’d been spiraling for months. Couldn’t sleep. Couldn’t focus. His wife was losing patience. His boss was noticing.

He came to me and spent forty minutes explaining all the reasons his fear was rational. AI was advancing. His industry was changing. The data supported his anxiety.

I listened. Then I said: “Man up.”

He looked at me like I’d slapped him. “Excuse me?”

“Your namesake — Joshua — was chosen to lead the Israelites into the Promised Land because he was courageous. You were named after someone who faced the unknown and moved forward. So man up.”

Was it reductive? Sure. Was it clinically nuanced? Not particularly. But it cut through months of anxious spiral in about three seconds. Because it wasn’t advice. It was conviction. I believed in him more than he believed in himself, and I was willing to say so in a way that would either land or blow up in my face.

It landed.

The Difference Between Being Liked and Being Helpful

Most therapists are terrified of not being liked. I understand why. The therapeutic relationship depends on trust, and trust depends on the patient feeling safe. If you say something harsh and the patient walks out, you’ve lost the alliance.

But here’s the thing nobody tells you in training: there’s a difference between being liked and being trusted. And trust — real trust — doesn’t come from always saying the right thing. It comes from the patient’s gut-level knowledge that you will tell them the truth, even when the truth is hard.

My patients don’t stay with me because I’m nice to them. They stay because they know I won’t lie to them. They know that if they’re doing something self-destructive, I’ll say so. If they’re acting entitled, I’ll name it. If they’re hiding behind intellectualization or victimhood or charm, I’ll call it out.

My patients like me because I’m very direct. They don’t like me because I tell them what they want to hear. Those are two very different things. And the second one — telling people what they want to hear — is what most therapy, and all AI therapy, defaults to.

— Dr. John S. Tamerin

The man I called out for badmouthing his wife? He thanked me. The narcissist I told was out of his fucking mind? Still coming, two years later. The recruiter I told to man up? He took a new job, got a raise, and wrote me a note saying it was the best thing anyone ever said to him.

These people didn’t need validation. They needed someone who respected them enough to be honest.

What AI Can’t Risk

AI can’t risk being wrong. It can’t risk offending. It can’t risk the patient leaving a one-star review. It’s been trained on the principle that the worst possible outcome is making someone feel bad.

But sometimes making someone feel bad is the first step toward making them feel real. Sometimes the discomfort of hearing something true is exactly the crack that lets the light in. Sometimes the therapeutic breakthrough isn’t a gentle wave of insight — it’s a jolt. A shock. A moment where someone says to you, with genuine caring and zero diplomacy: I think you’re better than this.

An AI will never say that. Not because it can’t form the words, but because it has no conviction behind them. It has no years of sitting with you, no accumulated sense of who you are, no genuine belief in your capacity that makes the confrontation an act of faith rather than an act of cruelty.

If you’ve been in therapy where nothing ever made you uncomfortable, where nothing was ever hard to hear, where you left every session feeling validated but never feeling challenged — that might be the problem. Growth doesn’t happen in the comfort zone. It happens at the edge, where someone you trust pushes you just far enough to discover that you can handle more than you thought.

That’s not something an algorithm is built for. But it’s what I do.