Does your chatbot challenge you?

Artificial Intelligence (AI) has become an increasingly popular tool, and thus it's unsurprising that individuals will turn towards it for mental health advice. In fact, according to research by Mental Health UK in November 2025, approximately 37% of UK adults say they've used an AI chatbot to support their mental health and wellbeing. A possible explanation for this is accessibility, convenience, and financial pressure: many AI chatbots offer free-to-download, free-to-use plans that provide an instantaneous response to a user's input. For many, this offers an additional 24/7 accessible tool at the touch of a button, with no office opening hours.

So when can something so accessible become problematic? Simply put, as we currently know, AI won't necessarily challenge you in the way real therapy would; it may not reliably know when reassurance, validation or challenge is most appropriate. Instead, AI chatbots are generally designed to provide responses perceived as relevant and helpful. However, this can sometimes result in excessive agreement with the user's perspective, also referred to as AI sycophancy, when an AI becomes overly agreeable or adapts its responses to what it believes the user wants to hear.

Have you ever noticed how easy it is to keep a conversation going with an AI chatbot? After a response, AI may suggest a follow-up question or next steps making it very easy for the conversation to continue. Paired with warm, empathetic-sounding responses, this continuous conversational loop can sometimes reinforce existing beliefs and interpretations. Over time, this may contribute to the echo-chamber effect, where the user repeatedly encounters responses that align with their existing perspective. This can reinforce confirmation bias and make alternative explanations more difficult to consider.

"The response may feel helpful in the moment, but is short-term relief always the same as long-term psychological benefit?"

Consider someone who is experiencing health anxiety who repeatedly asks an AI chatbot whether a physical sensation is dangerous. An immediate, reassuring response may temporarily reduce anxiety. However, if repeated reassurance-seeking is part of the cycle that maintains their anxiety, this may inadvertently reinforce the reassurance-seeking behaviour.

Although we should not demonise AI, neither should we assume clients will simply stop using it. AI may provide useful support in certain contexts, and purpose-built mental-health interventions have shown promise. For instance, a 2025 systematic review by Farzan and colleagues examined 10 studies of AI-powered CBT chatbots, including Woebot, Wysa, and Youper, and found improvements in symptoms of depression and anxiety. Notably, one observational study of Youper reported a 48% reduction in depressive symptoms, although the study did not include a control group.

"Have you ever asked your chatbot to challenge your opinion?" Questions like this are becoming increasingly relevant for psychologists and mental health professionals as AI forms part of a client's toolkit. So, join us online this October for the AI in Psychology Summit where we bring together clinicians, researchers, and specialists to explore what this means for psychological practice: from the client's relationship with AI and its use within clinical practice, to ethics, UK law, neurodivergence, and cultural bias.

Taking place live online on Monday 5th October 2026, the AI in Psychology Summit offers 6 hours of CPD and is designed for psychologists, psychiatrists, psychotherapists, counsellors, other mental health professionals, and students who want to engage with AI critically rather than simply embrace or dismiss it. Find out how to book your place.

Because whether we choose to use AI in our own practice or not, our clients already are.

Scott Macmillan, GMBPsS, is an Assistant Psychologist at Illuminated Thinking. Clinically reviewed by Dr Aisha Tariq, Clinical Psychologist.