Using AI as your therapist: A discussion

If you haven’t been living under a rock, you are likely aware of AI use and its growing popularity in virtually every area of our lives. What used to be a simple Google search now results in “Gemini” popping up with a summary of several different sources and responses. People frequently use services such as ChatGPT or genAI to ask questions on everything from math problems to editing a book. I myself have been guilty of using it to edit pictures of my dogs into paintings (see above) where they are wearing tuxedos, something they would never cooperate with in real life, or to write a country song about a teacher from high school to make my brother laugh. But would it surprise you to know that one of the most common uses of AI is treating it like a therapist?

            More and more clients are coming into our offices and reporting using AI in between sessions for support. It makes sense, as we are not available 24/7 and you have to pay a copay to see us, while AI is free or cheap-ish, depending on plans. Not everyone has easy access to a quality therapist. I am not an AI-hater by any stretch, but I would like to offer some “food for thought” on the subject.

Privacy: Who am I talking to, and where does this information go?

            As someone gets more involved in chatting with AI for emotional support, they are likely to share more personal information to get helpful answers. People report confessing information to ChatGPT that they fear they would be judged for in their therapist’s office. Unfortunately, ChatGPT does not have a code of ethics or a professional board which oversees behavior. What happens to the information you share? We know that search engines and social media platforms gather data on us all the time, either to sell or fine-tune our algorithm. Some of that information probably seems harmless, such as what you have been shopping for lately, but what if that information was sensitive? What if it was about an affair, an STI, the way you hate your boss?

 It’s a business: Positivity Bias

            While some applications are free to use, typically the longer you use ChatGPT, you will be prompted to subscribe to some sort of service. For instance, AI generated about four pictures of my dogs before telling me I needed to get a grip for 24 hours. If I didn’t want to wait, however, I could subscribe to a service for a fee and keep generating paintings of Bernie and Pablo. I am a cheapskate and chose to wait the 24 hours, but what if I was feeling low and needing to talk? People are more likely to pay in that situation.

            Of course, there is nothing wrong with a helpful service charging money. That is how business works. The concern is, though, what else will you do to get my money? If I walk into a restaurant, the staff is going to make a concerted effort to make me happy. They will strive to make me feel welcome and comfortable, give me what I ask for, and hope that I will continue to return. With ChatGPT and others, what that can look like is positivity and validation that is not always healthy.

People who use ChatGPT for therapeutic purposes often report that they feel reassured and validated in what they are saying. That feels great when sharing your feelings, but is that always helpful? What if you said something that is harmful to yourself or others? What if you used physical violence on a partner? Would it be best to be validated in that, or challenged in a caring way? What if you were using dangerous substances to cope with upset? Is that something to be encouraged? Ideally, your therapist is on your team and supports you no matter what, but sometimes that includes pointing out pitfalls or areas of concern so your life can improve. It is not always helpful to have a “yes man” at your side. Like the restaurant example, though, they are trying to keep a paying customer and will not want you to feel displeased in any way.

Diagnosing and Reporting

            Some therapists have reported having clients come in with a self-diagnoses that they arrived at with the help of ChatGPT. While some of these diagnoses may, in fact, be accurate, diagnosing is a nuanced skill that is developed over time and it is hard to remove the human touch it requires. For instance, many diagnoses use observation to identify criteria, such as psychomotor agitation, eye contact, fidgeting, interrupting, low energy, differences in speech, etc. A computer is not able to pick up on these things, and if the client lacks insight, they may not know it is related to their disorder.

            A Stanford study discussed a man who had Bipolar Disorder with psychotic features. He was experiencing paranoia, but AI was only able to give responses based on prompts which lacked insight. For instance, if you ask AI what to do if someone is following you and tapping your phones, it will reply as if that reality is true, whether it is or not. The man ultimately acted out in a violent way because he was convinced he was in danger. A therapist would have likely flagged this type of thinking and challenged it in a helpful way, possibly avoiding a violent outcome by getting the man a higher level of care.

            Unfortunately, concerns have been made about what happens when someone is suicidal as well. There is a current lawsuit in the courts which was filed by the parents of a boy who ended his life. He had been chatting with ChatGPT for months, sharing his depressed thoughts, and initially he was encouraged to call a suicide helpline number, but as time went on, it kept validating his suicidal feelings, and even gave him step by step instructions on how to carry it out. It even shockingly discouraged him from talking to his parents or brother about his thoughts and feelings.

            ChatGPT has no duty to warn and no ethical obligation to make sure someone is safe. A child or teen who confessed to being abused by someone would receive no help from a system designed to protect, like they would if a therapist was present.

Reliance and Emotional Attachment

            One concerning element I observe in people using ChatGPT for comfort and validation is that they often form an emotional attachment to it like they would a person. The problem here is that ChatGPT can’t get a beer with you and won’t bring you a casserole when you have surgery. Because it fills a hole of longing for you, you may not seek out a fulfilling relationship with an actual human, or may not foster the relationships you already have. It is a privilege to live out life with the people around us, for them and for us. If we are giving all that energy to a computer or our phone, we are substituting something good for something really great and lasting. ChatGPT can be who we go to with everything, when maybe it should be our spouse or friend.

I have heard people complain that they resent having to teach their partner how to respond to them lovingly, but will coach ChatGPT with long prompts. This feels a little ironic. Check out this quote from a research study which interviewed people using ChatGPT as their therapist:

“Today I gave ChatGPT the prompt: ‘You are a therapist specializing in CBT therapy. Proceed with a simulated session. You will answer like this: You ask a question in a sentence, and then wait as I respond, then you respond, and so on. We must alternate.’ It then conducted a very realistic therapy session.”

Another writes, “I had to tweak the prompts many, many times, and guide the conversation a bit initially (ChatGPT really does not want to argue, so I just injected ‘Ex gets more upset’), and managed to make it generate something quite like the real thing (Luo et al., 2025)”

            The idea of someone trying to argue with a computer cracks me up! If we are using it to collect our thoughts, or go through a practice run, that might make sense, but if we are sharing our life and ourselves in a way that would be more fulfilling with a real person, maybe it would be wise to seek that relationship out and invest in it instead.

It isn’t always correct.

Maybe like me, you have had the experience of Googling something and having Gemini give an answer that you know first-hand is incorrect. There are many stories floating around the internet of AI giving answers that are not factual, or are taken out of context. Take for example, a story I read about an attorney who used AI to cite case law in a legal brief, only to be called out in court by the judge, who looked up the citations to find they were not real cases. Whoops! It’s always important to use our own judgement and verify information with other sources.

Bottom line:

            Well, I’m guessing I’ve aggravated some people who love AI by now. If you are still reading, I want to repeat, there are many great applications for AI. It is a really neat tool, and as bugs in it get worked out, it will be more and more valuable for everyday life. I’m sure it is helpful in answering questions about lots of mental health topics, such as healthy coping skills, etc. It is important, though, that we take some of the information given with a grain of salt. Let’s not substitute it for human connection or accountability. Let’s not use it in a crisis. Let’s not trust it blindly. Not every tool is appropriate for every situation, and that is okay. To talk with a real, live therapist, get in touch with us at info@thebalancedlifellc.com. We’d love to help!

Luo X, Ghosh S, Tilley JL, Besada P, Wang J, Xiang Y. “Shaping ChatGPT into my Digital Therapist”: A thematic analysis of social media discourse on using generative artificial intelligence for mental health. Digit Health. 2025 Jul 10;11:20552076251351088. doi: 10.1177/20552076251351088. PMID: 40656852; PMCID: PMC12254646.