Your AI Therapist is Stealing from You
Updated: Oct 28
Should people talk to AI about mental health? In the APA’s 2026 survey of more than 1,200 licensed psychologists, 77% said they had patients who had used AI, and more than a third reported patients using AI as an additional mental-health provider.
Why are people using AI as a therapist?
It's not hard to understand the appeal.
Pocket therapist on the go? Reassurance on demand? Straightforward answers about behavior in relationships? Sign me up, sounds great.
We're addicted to instant gratification, answers, and information. And it's becoming more evident that the obstacles to therapy feel harder to navigate than its knock-off sidekick. Copays, deductibles so high it's like you're paying out of pocket for the entire year for services. Why even have insurance?
Then there's the waiting. Your therapist isn't available when you want to message them. You're laying in bed wondering what went wrong in that argument, wondering what that friend thought of you when you were there, wondering how you're going to go from this moment to actually going to sleep.
But you know who is there? Me, my phone, and AI.
And when you ask AI about its opinion it says insightful things, things you hadn't considered. And it's there with you in the dark, reassuring you that you did your best, surely it must not be that thing you did was embarrassing, of course you have a right to feel that way in the argument.
What We're Actually Doing
It's not great. And I think you already know that deep down. Let's talk through it to let you know how exactly how bad it is:
You can't decide where to go for lunch with a friend. You know what you both like and where you'd like to meet but you're both indecisive. Neither one wants to make a choice and it be the wrong one, hurt the other person or make someone uncomfortable, so you ask AI what it thinks based on preferences, location, and the vibe. It comes up with a map with restaurants and reviews and the menu.
Now here's what we're saying actually to our own brains if we're being honest (and I think we should be if you ask me): I'm not willing or capable of making this choice. I'm too uncomfortable with the idea of making this choice given the discomfort of what it could offer my friend and I'd be responsible if I did make the choice. You know who knows better than me? AI. I'll let it make the decision. I get a little less responsibility, I can make the decision quicker, and I'm more comfortable.
Here's another situation: You and your partner get into a big fight. You feel like you're justified in feeling and thinking the way you are, and you don't think your partner would agree. You're feeling minimized, invisible, or misunderstood and you want to check in to see if there's a way to understand what's happening in the relationship dynamic so that it could improve over time. You ask AI about what happened and your thoughts and feelings. It gives you a validating response that considers your feelings, lets you know it's okay to want repair, and then gives you a virtual forehead kiss and lets you know you'll be alright.
What are we doing, friends.
This isn't a question because I hope you know what you're doing. We're saying we think AI has a better understanding of our partner than we do, that it is a better stand-in significant other who can satisfy the emotional needs we have, and that we should feel and think this way. You're giving up a lot there, and it's all through your lens - the narrative, the emotions, the thoughts, how you behaved. There are some fundamental issues here surrounding who the authority is on the relationship and in handing the reigns over. I think we're getting dangerously close to being in a throuple with the robot.
"But Chelsea, it doesn't always agree with me," you'll say. Well guess what? Conjure a time where it said you were dead wrong, that you were over-reacting, that you got it wrong and had a tone that invited defensiveness. Mmmhmm. It's pandering to you.
"But Chelsea I could choose to not use it if I really wanted to." Oh yeah? Well then do that. And do it often. Check in with yourself when you do it and see how uncomfy you are, how much you believe in your answers without checking it at all. If you're unbothered and comfy, proceed. If you start to notice something else brewing, know that you've been warned.
Mental Health Implications
There are a lot of things that might not be hot takes here, but they're worth repeating anyway. Here's the biggest thing as someone who talks about control (professionally) multiple times per week with client: giving up the control you do have because of discomfort and fear is a recipe for telling yourself that you're actually not capable. And anyone who knows Exposure Therapy knows you don't let the inner goblin (your fears), make decisions on your behalf. They don't represent you, and they certainly don't represent who you are as a person.
If every time something feels uncomfortable or you feel nervous, you avoid it, I promise you the takeaway will be believing in yourself a little less everytime that you can engage with that feeling or situation or person. But that's the wrong takeaway. We're not losing capability, only confidence in that capability. And that's what's happening when you ask AI.
Call me a hater if you want. But what you're losing in the interaction is your agency. We could say AI is stealing your confidence, but I feel it would be more accurate to say you're willfully giving it away.
Appropriate AI Applications
Could AI be used as a mental health tool? Where should its role end? Can it be relied upon to provide any level or psychotherapy or psychological treatment? What's so wrong with an AI counselor?
Real-Life Scenarios
I'll give you a couple of nightmare scenarios happening weekly.
Client with OCD joins session and says that they're afraid they've come into contact with some sort of illness that will make them throw up. They asked AI six times in the last week about food-borne illnesses, symptoms, common cold and other illnesses circulating now, and stopped themselves from asking a bunch more times than they would have wanted to ask. I'm excited to hear they resisted the urge to check more often, and then we discuss how the additional information they got trying to check continued the rumination cycle for several days after the initial thought popped into their head.
And so the pocket reassurance is something that we need to consider here. Few to no relationships are accessible for reassurance any hour of the day. I'll argue all day that delaying urges and compulsions in general is the secret ingredient in the special sauce. It's what builds tolerance and, quite frankly, it's real life.
New client comes in and says they think they have Autism and Avoidant Attachment after doing some "research" online. They eventually reveal that they were chatting with AI to determine symptoms. Client walks out with an OCD diagnosis and some perfectionism that shows up in their relationship. Well that's not the same, now is it. Intervention for Autism and Attachment Trauma are so incredibly different from ERP. Double yikes.
AI is pattern recognition software. It's being implemented widely for that very reason and makes itself useful in so many ways. AI's place in the mental health sphere sounds like a good idea, but it's not designed to ruffle your feathers and question the way you said a word (yes, even if you have it on the mode where you ask it questions back and forth and don't type it out). Could it get some things right? Yeah, totally. But at what cost to your sense of self?
AI's Role in Your Mental Health
If you can't stop or don't want to stop using AI, do me a solid. If you have a mental health diagnosis, tell it that. If you don't have one and think you might qualify for one, go figure that out (and I mean with a highly qualified therapist). If you're part of the cool kids club and you have an Anxiety Disorder, OCD, or related disorder and do ERP, tell it that too. Ask it to respond in a way which is in line with ERP treatment and which calls you out on your compulsive shenanigans.
Before sharing a diagnosis or anything personal, check the tool’s privacy policy and settings. A chatbot is not a confidential therapy session. Keep identifying details out, and talk with your therapist about what is safe to share.
If you're looking for other possible ways to use it with your mental health, I'm going to struggle to cosign that. And not just because APA has warned about unsafe chatbot interactions that encourage self-harm, including suicide - also because you did it before and you can do it now. Do it scared. Do it uncomfortable. Just please don't give up on doing it.
Sources and further reading
1. American Psychological Association. Patients are bringing AI to therapy: 2026 Chatbots and Mental Health Survey. — Source for the opening survey statistics.
2. American Psychological Association. Health advisory: Use of generative AI chatbots and wellness applications for mental health. — Guidance on privacy, safety, and why chatbots should not replace qualified mental health care.
3. Abrams, Z. (2026). Discussing AI use in therapy. American Psychological Association. — Clinical perspectives on reassurance, missing context, and discussing AI use with a therapist; also considers possible supportive uses.
4. Cheng, M., et al. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391, eaec8352. — Research on overly agreeable AI responses and users’ willingness to repair interpersonal conflict.
5. International OCD Foundation. Exposure and Response Prevention (ERP). — Background on facing feared situations without compulsions or avoidance. This is OCD treatment guidance, not a study of AI.

Comments