Should you keep using ChatGPT as your therapist?
The consistent validation generative AI chatbots gives to their users becomes increasingly dangerous as people in vulnerable positions keep turning to them for advice. In 2025, over 19% of adolescents and young adults reported using AI chatbots for mental health advice. The problem with relying on artificial intelligence for psychological help lies in how chatbots are designed to respond to users. Large language models (LLMs) such as ChatGPT learn and improve through human feedback. However, the responses best received by humans are often validating, agreeing with, and mirroring them. Because AI chatbots are shaped by what people reward, they mirror users back to themselves, which doesn't substitute the honest disagreement necessary when someone needs support.
AI chatbots are specifically trained and designed to agree with the user. A 2024 study found that five AI assistants showed consistent sycophancy – excessive praise and agreement – across several different tasks. It was also found that humans often prefer responses that match their views, sometimes leaning towards sycophantic answers over correct ones. When trained, LLMs are optimized to produce responses that sound instantly helpful and supportive rather than responses that challenge users’ views. This design leads AI models to mirror their users, especially in situations of requesting advice and validation. 91.7% of users who have sought advice from AI chatbots rated it as helpful. However, this perceived helpfulness may be only a result of sycophancy instead of the quality of the advice.
Validation is crucial for support, but it should not be the only part. In therapy, it is important to view validation as a bridge to understanding, but not the ultimate goal. Of course, patients are heard and feel seen, but they also want to change and improve as people. This balance is crucial in therapy, but chatbots fail to achieve it. When people are overvalidated in their beliefs, they can't challenge themselves and become unable to grow and improve. When someone's choices harm themselves or others, they need to be corrected. Challenging oneself is crucial for growth. Therapists can offer patients insight after building trust and making them feel validated. The balance between reassurance validation and challenging is not seen when chatbots are used as therapists.
People who use AI as therapists often do so because they can't find the same support elsewhere. A study found that 63.3% of people who used chatbots for mental health advice had not told anyone about using them. For some, mental health support may be very inaccessible and chatbots end up being the only true option. Despite its support, AI should only be used as a gateway to care, not a substitute. A chatbot that only agrees with the user doesn't offer the same support as a person who is willing to disagree.
The comfort in using AI for mental health advice is tempting, but it only gives minimal support and the illusion of actual help. When seeking advice, it's crucial to consider an opinion that disagrees.
Sources:
Comments