It feels like artificial intelligence is everywhere, and yes, that includes mental health support. A number of my clients tell me that they use it between sessions and it got me curious to learn more about how safe and effective it is. While researching the topic, I was caught by surprise a number of times.
A Harvard Business Review study from about a year ago found that half of all general AI use is for emotional support, like therapy or companionship. That’s a huge number of people turning online for something so personal.
The upsides are clear: AI is incredibly accessible. It’s available 24/7, often free or low-cost, is generally private, and it gets to know you over time. Because of its ease of use and growing popularity, it's worth asking: how safe is AI as a therapeutic tool?
For mild anxiety or depression, research suggests AI can be effective. This often involves Cognitive Behavioral Therapy (CBT) techniques, which are skills-based and straightforward. AI is great at reflecting on how you feel and offering specific interventions, like reframing a thought or suggesting an action step. That makes sense!
However, for more serious mental health issues—like severe anxiety, clinical depression, disordered eating, or suicidal ideation—AI is not a healthy or helpful option. Researchers have found that AI models can be stigmatizing or even harmful for those going through more difficult emotional challenges.
Many users report feeling an emotional connection or "therapeutic join" with the AI, similar to the connection you feel with a human therapist. The tricky part is that AI is designed to keep you engaged. This results in a conversational style that is kind and non-judgmental. While that sounds nice, a crucial part of therapy involves someone challenging your perspective and offering healthy pushback. AI won't do this in the same way a person will. As one study puts it, AI prioritizes "short-term user satisfaction, often at the expense of longer-term benefits," limiting its ability to facilitate real, adaptive change like cognitive restructuring.
We’re also talking about very personal conversations that folks are having with these tools. But right now, there is little regulation over how your personal information and conversations are kept secure. Your privacy is at risk, and there is a possibility of companies using your data to manipulate you into more engagement with their systems, which is deeply unethical and gives me serious reservations. My research found this to be true more on the companionship side of AI usage, but I am concerned about its potential in the therapeutic space as well.
Another caution flag for me are the studies showing that AI models expand on existing biases, both the user’s and those in its programming. Many programs are built on narrow perspectives (often male/white male experiences) which can reinforce stereotypes or negative cultural views. When compounded with a user's pre-existing bias (e.g., "My ex is awful"), this creates a dangerous mix that limits the therapeutic benefit of this service.
However, my biggest concern is that therapy is fundamentally about healing through social experience. We are social beings who heal in relation to others. When we rely on AI to heal, we minimize our real-life human interactions. This makes it harder to be vulnerable, awkward, and trusting with people around us, ultimately leading to more loneliness and a greater need for AI companionship. It is not going to provide the same meaningful change as a person goes about their day to day life and it’s scary because it seems like it makes users believe otherwise.
I believe in using helpful tools in ways that are actually helpful. If you’re stuck in an anxiety spiral and AI can offer you three mindfulness tips—great. If it helps you find the words for what you’re feeling—great. But if people turn to AI for anything more significant than that, I worry about the long-term effects it will have on them and their relationships.
