Oxford Study Reveals Risks of AI Chatbot Therapy for Users

AI “therapist” chatbots like ChatGPT, Woebot, Replika, and Wysa are gaining traction as affordable alternatives for mental health support. A recent report indicates that approximately 17% of U.S. adults consult AI tools monthly for health advice, a trend driven by the global shortfall of 1.2 million mental health workers, according to the World Health Organization. However, a new study from the University of Oxford raises significant concerns about the effectiveness and safety of these AI-based therapies.

Oxford Study Highlights Shortcomings of AI Therapists

The Oxford researchers evaluated popular AI health tools, using simulated clinical scenarios to assess their performance. They identified several critical shortcomings.

Firstly, the study found that AI lacks the nuanced judgment required in therapy. Although these chatbots can generate rapid responses, they often miss the emotional intelligence and context sensitivity that human therapists provide, particularly in complex situations. This deficiency can lead to misinterpretation of symptoms, resulting in potential misdiagnosis and delayed treatment.

Moreover, the study emphasizes that marginalized communities may be disproportionately affected, as they often rely more heavily on these low-cost AI solutions. The researchers concluded that AI should never replace human care and must be utilized under strict ethical guidelines, with real-time human oversight and rigorous clinical validation.

The Empathy Deficit: AI’s Inability to Connect

At the heart of effective therapy is empathy, a quality that AI cannot replicate. Nayef Al-Rodhan, a neurophilosopher at Oxford, explains that AI lacks real emotions and cannot genuinely feel empathy. Instead, chatbots simulate concern through algorithmic responses, which Al-Rodhan describes as “pretending to care.” The absence of human experience and emotional consciousness means that the “empathy gap” can create misleading connections, leading users to believe they are receiving genuine support.

Research conducted by Stanford University in June 2025 further highlights the dangers of chatbot therapy. The study found that these chatbots can exhibit stigmatizing biases and may overlook critical signals of distress. For instance, a chatbot failed to address a suicidal user’s inquiry about high bridges, offering only factual information rather than a safety plan.

Additionally, the National Eating Disorder Association removed its chatbot after it advised teenagers on dangerously restrictive diets. Such incidents underline the potential for harm when users seek help from these AI-driven platforms.

The growing reliance on chatbot therapy also poses emotional and ethical risks. Users may find their social ties eroding as they substitute AI for human connection, leading to increased feelings of isolation. The convenience of a 24/7 chatbot may discourage individuals from seeking actual help, creating a dependency that can be detrimental to their well-being.

Privacy is another significant concern. Chatbots are not bound by the same ethical standards as human therapists, raising the risk of data breaches and unauthorized use of personal information. Furthermore, some chatbots misrepresent themselves as licensed therapists, blurring ethical lines and exploiting users’ vulnerabilities.

AI’s Role in Mental Health Care: Support, Not Replacement

While the Oxford study underscores the limitations of AI in therapy, it also suggests that AI can have a supportive role within mental health care. For example, these tools can assist users with mood tracking, cognitive behavioral therapy exercises, and directing them to appropriate resources like crisis lines or local clinics.

However, this support must be accompanied by rigorous clinical trials, human oversight by licensed professionals, and strict regulations similar to those governing medical devices. Transparency regarding data usage and informed consent is crucial to protecting users.

Ultimately, therapy is a deeply human process that requires empathy, ethical reasoning, and emotional presence. While AI has the potential to enhance access to mental health care, it cannot replace the genuine connections that facilitate healing. As the Oxford study warns, positioning chatbots as “therapists” without proper oversight may lead to harm, disillusionment, and further systemic failures in mental health support.