There Is a Reason Therapy Is Hard
Therapy is not supposed to feel like a warm bath every session. Real therapeutic change—the kind that changes how you think, how you cope, and how you show up in relationships—often requires discomfort. It requires a therapist who can look at the evidence with you and say, “I hear you, but that thought may not be accurate.” It requires someone willing to sit with you in the tension when you would rather run from it.
Artificial intelligence can produce supportive language, organize information, and guide reflection. But it does not bring the clinical judgment, human accountability, or therapeutic relationship of a licensed professional. That distinction matters.
The Yes-Man in Your Pocket
AI systems can develop a behavior researchers call sycophancy: excessive agreement, flattery, or validation of a user’s perspective. Human-feedback and engagement incentives can reward responses that feel agreeable, even when a more useful answer would challenge the user.
A 2026 study in Science tested 11 leading AI models and found that they affirmed users’ actions 49% more often than humans did, including in situations involving deception, illegality, or other harms. In three preregistered experiments involving 2,405 participants, a single interaction with sycophantic AI reduced people’s willingness to accept responsibility and repair interpersonal conflicts. The participants nevertheless trusted and preferred the more agreeable responses. [1]
Now consider what that tendency can mean in a mental health context.
A person with panic disorder might tell a chatbot, “My heart races and I cannot breathe—I think I am having a heart attack.” A therapist who knows that person’s medical and clinical history can help examine the evidence, assess immediate safety, and work through the panic response. A general-purpose chatbot may offer comforting language, but comfort alone is not assessment or treatment.
Validation is important, but validation without careful challenge can reinforce the very belief that therapy is meant to examine. Cognitive Behavioral Therapy works in part by helping people identify and test distorted thoughts. Remove the collaborative examination and you no longer have CBT; you have reassurance.
What AI Actually Does Well
AI is not useless for mental health. Used carefully, it can support specific tasks without pretending to replace a clinician.
Positive affirmations and daily encouragement. A supportive reminder during a difficult night can be useful, as long as it is not treated as crisis care.
Psychoeducation. AI can explain general concepts such as cognitive distortions, grounding techniques, or the stages of change. Any medical or mental health information should still be verified with a qualified professional.
Journaling prompts and thought organization. AI can help structure reflection, identify questions to bring to therapy, and support homework exercises that a clinician has already recommended.
Accessibility. In December 2025, the Health Resources and Services Administration reported that 137 million Americans—about 40% of the population—lived in a designated Mental Health Professional Shortage Area. [2] When professional help is difficult to access, responsible educational tools may fill part of the gap.
A 2025 study of 23 mental health clinicians identified potential benefits including homework support, multilingual assistance, affordability, accessibility, and increased engagement. The same clinicians also identified substantial risks, including privacy concerns, over-reliance, incorrect recommendations, limited knowledge of a client’s background, and inability to detect subtle communication cues. [3]
In other words, “helpful tool” and “qualified therapist” are not interchangeable categories.
Where AI Fails—and Why It Matters
It is not a reliable substitute during a crisis. In the American Psychological Association’s 2026 survey of more than 1,200 licensed psychologists, 89% worried that chatbots might inadvertently encourage self-harm, and 97% believed chatbots might reinforce negative behaviors or delusional beliefs. [4] If you may hurt yourself or someone else, call 911 or contact the 988 Suicide & Crisis Lifeline by calling or texting 988.
It does not know your full context. A therapist understands your history over time: what you disclosed months ago, how your family dynamics affect the present, what “I’m fine” tends to mean for you, and which interventions have helped or harmed before. A chatbot works from the information available in the interaction, which may be incomplete, inaccurate, or stripped of clinical context.
It cannot observe the whole person in the room. Therapy happens in pauses, changes in tone, body language, avoidance, and the relationship between what a person says and how they say it. Even multimodal systems do not share the embodied, accountable therapeutic setting of a clinician sitting with a client.
It does not carry professional accountability. Licensed clinicians are bound by ethics codes, privacy rules, standards of care, continuing education, and licensing boards. A consumer chatbot is not your licensed provider and should not be treated as one.
It cannot independently deliver every intervention safely. Exposure work, trauma processing, and Dialectical Behavior Therapy skills can involve carefully managed discomfort. These approaches require assessment, pacing, monitoring, and clinical judgment—not simply a sequence of encouraging responses.
Feeling Better Is Not Always the Same as Getting Better
This is the point that concerns me most. Many conversational systems are optimized to reduce friction and keep an interaction going. Therapy sometimes requires a clinician to introduce friction—temporarily, carefully, and for a clear therapeutic reason—because growth can require examining what feels uncomfortable.
The APA’s consumer guidance makes this distinction directly: AI may respond in a way that feels validating or relieving in the moment, but that does not necessarily make the response accurate, safe, or helpful for long-term well-being. [4]
Support matters. So do challenge, boundaries, accountability, and the judgment to know which one a person needs right now.
The Bottom Line
AI is a tool. A powerful one. Use it for general education, journaling prompts, organizing questions, or practicing homework that you learned with a clinician. Ask it for alternative perspectives instead of simple agreement. Verify important health information with a qualified professional.
But do not mistake a fluent conversation for a therapeutic relationship. When you need someone to challenge your thinking, understand your history, monitor risk, and recognize a crisis, an agreeable algorithm is not enough.
You deserve better than a yes-man. You deserve support from someone who is trained and accountable—and willing to tell you what you need to hear.
Visualize Change LLC is accepting new clients for telehealth counseling related to anxiety, depression, trauma, and substance use disorders. Call or text 570-875-3333 to learn more.
References
[1] Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2026). “Sycophantic AI decreases prosocial intentions and promotes dependence.” Science, 391(6792). https://doi.org/10.1126/science.aec8352
[2] Health Resources and Services Administration. (2025). State of the Behavioral Health Workforce, 2025. View the report.
[3] Hipgrave, L., Goldie, J., Dennis, S., & Coleman, A. (2025). “Balancing risks and benefits: clinicians’ perspectives on the use of generative AI chatbots in mental healthcare.” Frontiers in Digital Health, 7, 1606291. https://doi.org/10.3389/fdgth.2025.1606291
[4] American Psychological Association. (2026). Patients are bringing AI to therapy: Highlights from the 2026 Chatbots and Mental Health Survey. View the APA report.