ok so I’ve been getting this question a lot lately — from readers, from Bill (my 67-year-old stepdad who just asked me whether “the AI app” his doctor mentioned is the same as “the one you write about”), and honestly from my own Woebot-using self when I’m trying to explain what I’m actually doing at 11pm on a Tuesday.
The short answer is no. They’re not the same. And conflating them causes real harm — not theoretical harm, not harm-that-might-happen-someday, but the kind where someone downloads Replika thinking it’s going to help with their anxiety disorder, or refuses to try Woebot because they’ve heard that “AI therapy” is a gimmick.
Let me try to unsort this.
Two Categories. Very Different Things.
The first category is AI therapeutic tools. I’m going to name three specifically because the generic term “AI therapy app” lumps together things that shouldn’t be lumped.
Woebot was built out of Stanford in 2017 by Alison Darcy, a clinical psychologist. It’s based on Cognitive Behavioral Therapy — specifically, the kind of CBT that’s been studied in randomized controlled trials. The app has published research in peer-reviewed journals. It tracks your mood. It walks you through CBT exercises. The penguin mascot is admittedly very earnest. But there’s a meaningful clinical framework underneath.
Wysa is similar in spirit — structured therapeutic frameworks, not just a chat interface. It uses CBT, DBT, and mindfulness-based approaches. Not replacing a therapist. But not just a chatbot that says “that sounds hard.”
Sonia is the newest of the three and worth knowing about. Y Combinator-backed, matched voice sessions specifically for Generalized Anxiety Disorder. They’ve published clinical data. They’re targeting a specific diagnosis with a specific modality. That’s a more honest product positioning than most.
These are clinical tools. Peer-reviewed, condition-specific, built by people who understand what they’re treating and what they’re not.
The second category is social companions. Replika, Character.AI, Pi. These are not trying to treat a condition. Replika explicitly describes itself as “an AI friend who is always there for you.” Character.AI lets you build a persona or choose from millions. Pi is warm and reflective and good at conversation.
None of these are bad. I’m not here to tell you not to use them. Loneliness is real, human connection is hard, and sometimes you need something to talk to at midnight that won’t be weird about it in the morning.
But they’re not therapy tools. And if you’re using them hoping to treat anxiety, depression, trauma, or any actual clinical condition — I’d gently suggest that’s not what they’re built for and the data doesn’t support it.
The Conflation Problem
Here’s why the conflation matters practically. The APA released guidance in 2024 on AI in clinical practice. The framing was careful: AI tools can augment clinical care, particularly for administrative functions and as adjuncts to between-session work, provided the clinician retains responsibility and there’s meaningful informed consent.
That guidance was written for clinicians. But it reflects something important for users too: the question isn’t “is AI good or bad for mental health?” The question is “what is this specific tool designed to do, and does that match what I need?”
Stade, Stirman, and colleagues published research in npj Mental Health Research in 2024 on LLMs in behavioral healthcare. The nuanced finding — which tends to get lost in the “AI therapy is coming for your therapist” discourse — was that the evidence base looks better for structured, condition-specific applications than for general-purpose companion apps. Which is exactly what you’d expect. Tools built for specific clinical purposes with clinical validation perform better clinically than tools built for general conversation.
I don’t have an easy answer for whether that means Woebot will replace your therapist. My $250-an-hour therapist uses Claude for SOAP notes, which I found out by asking, and I’m still figuring out how I feel about that. I’m not sure the therapy-replacement question is the right one anyway. The better question is: what are you actually trying to accomplish?
Three Questions to Ask Yourself
I want to be clear that these are not diagnostic criteria. If you’re in crisis or actively struggling with your mental health, please talk to an actual clinician. What I’m offering here is a decision framework for people who are genuinely confused about which type of tool might fit what they’re looking for.
Question 1: Are you managing a specific condition or symptom cluster?
If you have diagnosed anxiety, depression, OCD, or another condition — or if you strongly suspect you do — you’re in the territory where clinical tools matter. Woebot, Wysa, and Sonia exist specifically for this. They’re not replacements for a therapist, but they’re built on the same evidence base your therapist uses. The companion category wasn’t designed for this and the research doesn’t support using it this way.
Question 2: Are you lonely, or do you want to feel heard?
This is a legitimate need and the companion category is designed for it. If what you’re looking for is someone to talk to, a warm presence, a conversation that doesn’t require you to perform okayness — Replika, Pi, and similar tools do this well. Just don’t mistake feeling heard by an AI for making progress on underlying loneliness, which requires actual human connection. Those are different things.
Question 3: Is this in addition to human care, or instead of it?
The hardest one. If you’re using any of these tools as a substitute for therapy you can’t access or can’t afford, I understand that, and these tools are probably doing some good. But if you’re using them as a substitute for therapy you’re avoiding because it’s hard or expensive or you’re not sure it will work — that’s worth looking at honestly. Both categories of AI mental health tool work best as supplements, not replacements.
What I Actually Use, and Why
I downloaded Woebot on a Tuesday in early May at 1:42 in the afternoon. I was in between therapy sessions — my therapist and I moved to every other week, which I suggested and she agreed to — and I was having a hard week and didn’t want to wait ten days.
What I found was that the CBT check-ins were useful in a way I didn’t expect. Not because the AI understood me — it doesn’t, not really — but because going through the exercise of naming what I was thinking, identifying the distortion, and choosing a more accurate thought is a skill that requires practice, and the Woebot format made me practice it. Like how doing the same physical therapy exercises at home between sessions makes the sessions work better.
I’m not going to tell you it replaced my therapist. I’m not going to tell you it didn’t help. Honestly, I don’t know yet. I’ve been doing this for a few weeks and I’m trying to stay honest about what’s actually changing versus what just feels soothing in the moment.
That kind of uncertainty is, I think, the right posture for all of this. These tools are new. The research is early. The marketing is way ahead of the evidence in both categories. If you go in knowing what you’re looking for and honest about whether it’s working, you’re in better shape than most.
One More Thing
If you’re a man reading this trying to figure out whether to try any of this — I’d say the biggest obstacle isn’t usually picking the wrong tool. It’s not picking any tool because doing so feels like admitting something is wrong.
Nothing’s wrong with using a structured CBT app to work on anxiety. Nothing’s wrong with wanting something to talk to. The conflation between these two categories is a product-marketing problem, not a character flaw problem. Figure out what you’re looking for, find the tool built for that purpose, and give it an honest try.
The decision is yours. I just want you to make it with accurate information, which is a harder thing to come by in this space than it should be.