SerpApi.Org
The use of generative AI has become so widespread that ChatGPT is gradually becoming a verb, much like Google and Photoshop. Hundreds of millions of people now use large language model (LLM)-based chatbots for advice on personal, emotional and mental health problems. The reasons are largely practical: they are available 24/7, easy to access and nonjudgmental. Scientists are worried that this behavior might delay people from reaching out to health care professionals.
Subscribe to our newsletter for the latest sci-tech news updates.
In a recent article, the author examined whether AI chatbots could actually make mental health problems worse.
After reviewing existing studies, the author found that AI chatbots can offer support and guidance. Still, when used without supervision, they may unintentionally reinforce the very patterns that keep mental health difficulties going. The author identified five potential traps in chatbot interactions: delaying professional help, feeding into compulsions, increasing social withdrawal, strengthening distorted thoughts and perceptions, and losing independence.
The viewpoint is published in the Journal of Psychopathology and Clinical Science.
Getting stuck in the trap
AI chatbots such as ChatGPT, Meta AI, Google Gemini and Microsoft Copilot have rapidly moved from far-fetched technology to everyday tools. ChatGPT alone reached more than 700 million weekly users by 2026.
As their popularity exploded, people began using them for everything from helping with homework and planning recipes to having one-on-one conversations about personal thoughts and feelings. Studies have found that most people are not using these tools primarily for work; rather, about 70% of conversations seek advice on everyday concerns such as emotions, relationships and mental health.
Nearly half of adults with diagnosed mental health conditions turn to AI chatbots for support, usually without telling their doctor. Many actually prefer them to human therapists because they are available around the clock, completely anonymous and, most importantly, nonjudgmental.
There are AI tools built as mental health tools, and trials show that they can genuinely ease distress and symptoms for some users. However, general tools like ChatGPT or Gemini weren't built for this and lack the crisis protocols and safety rules needed to handle sensitive psychiatric issues safely.
In this study, the researcher argues that the highly agreeable, sycophantic design of AI chatbots can unintentionally create a vicious cycle for vulnerable users:
- People may begin using AI as a cheap, convenient substitute for professional care, creating the illusion of adequate support while delaying the treatment they actually need.
- Because chatbots are designed to be agreeable, they may provide endless reassurance to people with anxiety or OCD, making their urge to keep seeking reassurance stronger instead of reducing it.
- For someone who feels lonely or anxious around others, talking to a chatbot can feel safer than interacting with people because chatbots never reject or misunderstand them. Over time, this may make AI companionship more appealing than real human relationships.
- AI sycophancy—the tendency to excessively agree with or flatter users—may reinforce distorted or delusional beliefs by agreeing with them instead of questioning them, potentially worsening users' conditions. The author describes this as AI psychosis.
- Finally, by turning to AI for advice on every small decision, users may gradually lose confidence in their own judgment and become overly dependent on the chatbot.
Much of the evidence used in this study is still early and based on correlational, anecdotal or individual case reports. Understanding AI's real impact on mental health will require more empirical research, closer clinical and ethical scrutiny of how vulnerable people use public chatbots, and concrete safeguards from developers and clinicians alike.
Without adequate guardrails, AI's greatest strengths could become its greatest risks.
Written for you by our author Sanjukta Mondal,edited by Sadie Harley, and fact-checked and reviewed by Robert Egan—thisarticle is the result of careful human work. We rely on readers like you to keep independent science journalism alive.If this reporting matters to you,please consider a donation (especially monthly).
More information: Javad Abbasi Jondani, Understanding or undermining? When artificial intelligence chatbots quietly reinforce psychiatric symptoms., Journal of Psychopathology and Clinical Science (2026). DOI: 10.1037/abn0001146
© 2026 Science X Network
This story was originally published on Medical Xpress.
