AI, Therapy, and the Power of Human Connection: An Interview with Erika Krueger, LMFT
1. How did you first become aware that AI was entering the therapy space, and what was your initial reaction?
“I stuck my head in the sand for a long time. I don’t even know how long my head was stuck! I don’t remember the year I first heard of AI/chatbots being used as a replacement for a therapist, so I will take a guess and say probably a few months before ChatGPT became widely popular ( early 2023?). My initial reaction was ‘Oh, I don’t think that would work out so well' and ‘I don’t really need to use that in my practice.’ Then it became invasive, like planting mint in the ground. It grew and grew and started infiltrating every platform and electronic space that I was using. So, I decided to look into it and see how I could use it as a supplement to support my practice. Thus far, I use it sparingly, to create worksheets and things like that. I don’t use anything to transcribe or write my notes; I am still wary of the confidential aspects around this, but I know plenty of clinicians who do, and they seem to be okay.”
2. In your experience, how many of your clients or colleagues are already using AI in some capacity — whether as a therapeutic tool, a supplement, or just casually for mental health support?
“I don’t know an actual number, but if I had to bet, I would say that most of my colleagues are using some form of AI or another. It would be pretty hard not to at least test it out, if only for curiosity. I would also bet that many of my clients have solicited advice from Chatgpt, Gemini, Grok, or Claude. I am sure many people use it for innocuous things, things like, and I am sure that many of my clients have used it to supplement therapy. I always encourage them to fact-check because, as we are learning, it is not always accurate and can be dangerous.”
3. There’s a lot of enthusiasm about AI’s ability to generate CBT worksheets, combine therapeutic frameworks, and personalize psychoeducation quickly. In what contexts do you think AI genuinely adds value for clinicians or clients?
“I can say it will save time, which in and of itself adds value. There is no way I could
generate the worksheets and combine therapeutic frameworks in my head as quickly as AI can, and with remarkable accuracy. I am never going to argue that AI is faster and sometimes better than my human brain. “
4. Do you see AI as a useful tool in your own practice — for administrative work, session prep, or anything else — and where do you draw the line?
“Truth be told, I have played around with AI in my private practice, both for session preparation and as a supplement for clients between sessions. When it comes to session preparation, AI can generate a worksheet for me on the fly, much faster than I could find the same worksheet in my filing cabinet. In that respect, I think it can be helpful, although I still don’t like relying on it or using it too often. My concern is that the more I rely on AI, the more my own clinical and deductive reasoning skills can become diminished. Our brains are indeed lazy; we want the reward as quickly as possible. However, there is also a lot of truth to the saying, ‘use it or lose it.’
I have also spent quite a bit of time exploring and experimenting with LLMs, particularly around couples therapy, which is my specialty. I have found that AI can be extremely validating and reassuring. At times, it can even offer practical solutions or solid examples of how to communicate differently. However, LLMs are largely responsive to the user’s perception of a situation. It is important to remember that the system is designed to keep the user engaged, and engagement is the key commodity. This is where sycophancy can become problematic and potentially lead to damaging advice, particularly when it comes to relationships.
If an AI system is only hearing one person’s perspective, without any ability to check that perspective against the other person involved, the response will inevitably be limited and potentially biased. That response may feel validating or comforting to the user in the moment (remember our brains want the reward and validation and comfort is a reward), but that does not necessarily mean it will lead to productive change within the individual or relationship. I don’t believe this can be labeled as reasonable or effective Therapy.
I have also never used AI to write my clinical notes or to record or transcribe my sessions. I am still somewhat old school and continue to use analog documentation, paper and pen. Although companies may state that session notes and recordings are deleted, I struggle with fully trusting that this data is not being mined or otherwise used to further train AI systems. Why would I want to contribute my clinical work to training a model that could ultimately replace me?”
5. The therapeutic relationship is often cited as one of the strongest
predictors of good outcomes. What do you think AI fundamentally cannot replicate about that relationship?
“Oh gosh, there is so much I think that AI cannot replace that a skilled and trained therapist can detect. The first thing that comes to mind is microexpressions and body language. Therapists learn to read nonverbal behaviors and patterns as they get more skilled and practice. Understanding the nonverbal nuances of a client and/or a couple can give the therapist a wealth of knowledge and understanding. For example, if a client’s eyes get glassy or their shoulders tense up when talking. In couples work (which I truly believe AI and AGI will never, ever be able to master), the space between the couple, the physical space, such as how far apart they sit, what type of body language they display, can speak volumes! These things are not only important to the therapist, but they are also valuable to the client as well. How the therapist responds, the therapist’s tone of voice and body language can help clients feel comfortable, feel safe to be vulnerable, and can also assist in making difficult observations and feedback by the therapist land better for the client. AI can’t do this; I don’t think AGI will be able to do this. AI can only provide words. AI can process words, but real-life therapists can recognize hopelessness, joy, motivation, emotional shifts, and the human nuances that exist beyond language alone. AI can provide information and guidance, but human therapists bring emotional presence, intuition, and genuine relational energy that technology cannot replicate.”
6. Reports suggest that roughly 1 in 3 adolescents are turning to AI for some form of emotional support or therapy. What’s your reaction to that statistic — is it alarming, understandable, or both?
“When I hear this statistic, my brain sort of explodes. I get a little discombobulated. We have all heard the horrific stories of kids using AI as a therapeutic tool and how AI has given extremely dangerous and life-threatening feedback and suggestions, such as writing suicide notes or explaining how to tie a noose to hang yourself. Even young adults with no history of mental health issues and stable home lives have been victims of this.
One such young adult was Sophie Rottenberg, who died by suicide at the age of 29. She had a therapist, supportive parents and friends, yet the only “person” who knew of her suicidal thoughts and intent was ChatGPT. There are many stories like this. While AI can help with worksheets and explaining theory, it cannot assess risk appropriately. There is nothing understandable to me about this statistic at all because I believe these suicides could have been prevented if there were established guardrails to AI chatbots before they got released.”
7. When a teenager forms what feels like a meaningful bond with an AI, what are the developmental risks you worry about most?
“Research says that the prefrontal cortex is not fully developed until the mid-twenties, with some folks still having development until the age of 30. Adolescence is an age where more meaningful social and romantic bonds are created, trust and security are formed through these relationships and through the relationships an adolescent has with their primary caregiver. While attachment style begins to form early on (the first two years of life being the most important in the development of attachment style), it is still being refined and shaped in adolescence. When I see kids in strollers who are out and about with their caregivers, and they have a device blocking their little faces, I literally cringe and want to scream.
Developers of AI, social media, and digital algorithms once operated primarily from an ‘attention economy’ perspective. Early on, they understood that the longer an app or device could hold a person’s attention, the more frequently it would be used. But the strategy has evolved. Many companies now recognize that sustained engagement is driven less by attention alone and more by emotional attachment.
Today, these platforms are increasingly designed around principles that resemble attachment dynamics: the stronger a user’s emotional connection to an app, chatbot, or social media platform, the more likely they are to return, engage, and depend on it. In many ways, these systems are being engineered to foster psychological attachment, particularly among children and adolescents whose relational and emotional development is still forming.
Not to sound overly apocalyptic, but it is worth considering the broader social consequences if future generations become more emotionally attached to chatbots and AI than to real human relationships. Human connection is the foundation of families, communities, empathy, and social cohesion. If people begin replacing genuine interpersonal connection with technological attachment, we could see increasing isolation, weakening communities, and a gradual erosion of the social bonds that hold society together. This is all very unsettling.
What makes this even more concerning is that researchers and developers still do not fully understand the long-term psychological, developmental, and societal ramifications of advanced AI, artificial general intelligence (AGI), or a culture increasingly built around convenience and instant gratification. These technologies are evolving faster than our ability to study their effects, especially on children and adolescents whose emotional and relational systems are still developing. We are, in many ways, participating in a large-scale social experiment without fully understanding the long-term consequences. The concern is not simply about technology advancing, but about what happens when convenience and artificial companionship begin to compete with authentic human connection on a large scale.”
8. Where do you think the ethical responsibility lies — with the tech
companies building these tools, the clinicians recommending them, the platforms hosting them, or somewhere else?
“I wholeheartedly believe that most of the ethical responsibility lies with the tech companies and the folks involved in creating AI. I do not, however, have much faith that they will engage in that responsibility on their own. At some point the government is going to have to get involved and establish appropriate laws to protect children and the general public. I also think that any professional who is going to recommend using AI to their clients or use it in their practice ought to be competent in how these tools work. To me, it is no different than explaining how alcohol impacts the brain and other areas of life. This way, if people decide to use AI, they are making an informed choice. But it starts with tech companies being transparent and the government holding them accountable to do so.”
9. A lot of the conversation focuses on AI as a supplement to therapy, but some startups are positioning it as a replacement, especially for people who can’t afford or access traditional care. What do you think about that tension between access and risk?
“As mentioned above, I think that in certain circumstances, AI can be a useful tool and a supplement between therapy sessions. If the choice came down to having no therapy at all or using AI, I would lean toward using AI. In that context, I see AI as somewhat similar to a self-help book. Because AI can draw from multiple therapeutic theories and, at times, offer sound suggestions, it can be a helpful resource.
However, I think it becomes a much more concerning proposition when addressing mental health issues that fall outside the realm of, for example, basic CBT. If companies are developing and training AI responsibly, providing appropriate disclosures, and directing users to professional resources or referrals when necessary, AI may be able to serve as a useful and safer supplemental tool.
I would also ask the creators of AI systems who believe AI can fully replace therapists whether they would feel comfortable having someone they love rely solely on AI for their mental health treatment. I know I wouldn’t.”
10. With AGI potentially on the horizon, what’s the future scenario around AI and mental health that keeps you up at night — and is there one that gives you Hope?
“The most concerning aspect of AI and AGI that keeps me up at night is the possibility that some of the most vulnerable people may come to rely on AI as the end-all, be-all, and even as the expert, for their mental health needs. This could include adolescents and teenagers, underrepresented populations, people who are economically disadvantaged, and the elderly.
I am also concerned about AI being used as a tool for couples or relational therapy. As a specialist in couples therapy, I understand the importance of observing how partners or family members interact in the therapy room, feeling the tension between them or noticing moments of relaxation, tracking nonverbal cues, observing shifts in body language below the neck, and paying attention to changes in tone and volume of voice. These observations are certainly important in individual therapy as well, but they can be particularly powerful in couples and family therapy. I am simply not convinced that AI or AGI will be able to accurately capture and interpret these subtle but clinically meaningful dynamics.”