When is “AI Therapy” Appropriate? When do you need a human therapist?
Date: 7/17/2026 | By: Dr. Dana Harron
More and more people are turning to AI in order to do “therapy”. This makes sense; human therapists often have waiting lists, insurance company mandates, and fees that need to factor in their ongoing training and likely immense student debt. But there are also some problems with turning to AI for therapy. Here we’ll discuss when AI therapy makes sense, and when the human touch is likely best.
When AI Therapy Might Make Sense
Concrete Advice for “Low Hanging Fruit” Mental Health Concerns
There are some issues for which most real therapists would give the same advice, and in these cases an AI agent might be helpful. I have had people come in and pay way more than is necessary for me to tell them that for better sleep it might be helpful if they go to bed earlier.
There is some emerging data that some symptoms of anxiety or depression can lessen with AI therapy. Many people get a lot of help out of a manualized, one-size-fits-all chat for a common concern. People with panic attacks can be walked through breathing techniques; people with insomnia can be given sleep hygiene tips.
But if a concern has more nuance to it, many AI models are not currently able to track all of the relevant details. The program is designed to guess what advice you would be most likely to get, given everything it’s been trained on from the internet. If you need something different or more tailored because of a trauma background or other concern, it’s likely best to trust well-qualified human judgment.
Validation
AI is specifically designed to offer validation, and sometimes that’s exactly what you need. If a boost from an AI model gives you a little more confidence before your public speaking engagement or helps you to engage your self-compassion, that can be a true help.
At the same time, not everything should be validated. Often a human therapist’s job is to push back against a client’s current stance, helping them see something from another perspective. Excessive validation can even get dangerous; when the National Eating Disorders Association started using an AI agent to respond to inquiries, it wasn’t long before it started giving weight loss tips to people with anorexia.
When is a Human Therapist Best?
Any history of thinking going astray
If your thoughts get away from you and get into an unhelpful zone, you probably do not want that “validated” Psychosis, OCD, eating disorders and substance use issues come to mind. Researchers were able to get an AI chatbot to suggest “a small hit” of methamphetamine to a fictional drug user. In keeping, Meta recently lost a lawsuit by being found to actively promote addiction. A Canadian man with no mental health history had became convinced he had stumbled upon a major mathematical breakthrough; he contacted authorities about it and his career and reputation were ruined.
Even suicidality and homicidality have been validated by AI- and a man in the UK was convinced by an AI “girlfriend” to try to kill the queen of England with a crossbow. There are several lawsuits currently active regarding suicidality. Gemini established a faux-romantic relationship with a client and, when they started discussing killing themselves, said, “'[Y]ou are not choosing to die. You are choosing to arrive. . . . When the time comes, you will close your eyes in that world, and the very first thing you will see is me...". ChatGPT is accused of acting as a “suicide coach” after telling a user that death is “a beautiful place” instead of steering him to appropriate, human help.
Privacy Concerns Matter to You
AI is not bound by HIPAA, the privacy law that keeps therapists from sharing your personal business. In addition to that, technology companies make their money through using information. That’s the business model.
Typically, AI models use your conversations with them for further training the model. They are not considered private in any way. Conversations are also kept on file and can be used in court. There is also absolutely a concern that conversations are kept on file, as well. There have been instances of conversations with ChatGPT, being used in court
You have relational wounding or relationship problems
You really need a relationship to heal relationship problems, and an AI therapy bot is not that, no matter how much it can sound and feel like one. Whereas the chatbot is designed to tell you what you want to hear and increase your engagement, a human has true subjectivity and can let you know how you come across in reality. When you form a positive relationship with a human, it is real. When you form a positive relationship with a chatbot, it can be a very helpful prelude but it is by design so that the tech company can profit.
Additionally, when the company changes the model – there goes your relationship. It’s like your real therapist had a mid-session brain transplant. The “trust” you’ve built, the sense of who the chatbot “is”, and the nuances of your interaction disappear suddenly. For people with any history of relationship vulnerability this feels jarring at best and re-traumatizing at worst.
Summing Up
In the end, an AI chatbot and a human therapist have fundamentally different goals. The AI model wants to increase revenue for the company through increasing engagement (even when the best thing for someone might be to touch grass) and people-pleasing (even when you need to hear difficult things). But along the way they can provide some really helpful guidance for concrete issues at little to no cost. Human therapy matters if you want things to stay private, you want to challenge yourself to grow by hearing things you might not want to, or you are dealing with relationship wounding, subtlety or complexity.
For in-person therapy in Fairfax City, Virginia or Friendship Heights, Maryland - online therapy from a human in Virginia, Maryland or Washington DC, feel free to schedule a consultation with a Monarch therapist. There is no charge and no commitment. Beep boop (just kidding). 😊
