The relationship between OCD and AI dependency is, on its face, a story about convenience: a tool that is always there, never tired, never irritated, never unavailable at 2am when the intrusive thoughts arrive. But it is also, increasingly, a story about the particular way a machine can make things quietly worse while appearing to make them better.
Stylist has been exploring what daily life looks like for people with OCD who have come to rely on AI chatbots, and the picture that emerges is one many clinicians will find uncomfortable reading. The appeal is not difficult to understand. When reassurance is the compulsion, a chatbot is a near-perfect enabler. It answers every question. It does not sigh. It does not remember that you asked the same thing four hours ago. It simply responds, and in doing so, feeds the very cycle it feels as though it is resolving.
Why OCD and AI dependency is a clinical concern, not just a digital habit
According to TreatMyOCD, ChatGPT can unintentionally reinforce OCD compulsions, including reassurance-seeking and repeated checking, precisely because it is available around the clock and never sets limits. That structural quality, the one that makes it feel like a lifeline, is the same quality that makes it dangerous for someone whose disorder is driven by the need for certainty. There are no friction points. There is no waiting room. The boundary that a therapist holds, the one that says: no, I will not answer that question again because answering it is harming you, simply does not exist.
People with OCD are turning to AI-based tools for support, information and symptom management in growing numbers, and there is genuine clinical potential in some of what these tools offer. The National Association of Addiction Treatment Providers notes that AI can potentially enhance psychoeducation, treatment planning and Exposure and Response Prevention (ERP) therapy, which is the gold-standard clinical approach to OCD. ERP works by helping patients tolerate uncertainty without performing compulsions: in supervised, structured use, a well-designed AI tool could, in theory, support that process between sessions.
The word “supervised” is doing a great deal of work in that sentence.
The gap between clinical promise and what actually happens at 2am
Because what actually happens, for many people, is something closer to the opposite. The Journal of Nursing and Patient Care has found that AI systems may misinterpret compulsions as normal behaviour, or fail to provide personalised treatment recommendations. A chatbot offered reassurance about contamination fears is not delivering a therapeutic intervention. It is delivering the compulsion itself, in clean, authoritative language, with no memory of having done so forty minutes before.
This is where the lived experience Stylist describes becomes something beyond an anecdote. The person asking a chatbot everything because it cannot say no is not describing a quirky digital habit. They are describing a compulsion loop, mediated by a tool that was not designed for them and does not know what it is doing to them. The 24/7 availability that makes large language models genuinely useful for most queries is, for someone with OCD, a structural risk. The machine does not know the difference between a question that deserves an answer and a question that should be met with silence.
The National Association of Addiction Treatment Providers is explicit that AI in OCD treatment introduces significant risks alongside its clinical promise, and that the distinction matters. Potential is not delivery. A hammer is not a surgical instrument simply because both are tools. The question for clinicians, developers and the people caught in between is whether the design of these platforms will ever catch up with the complexity of the conditions they are being used to manage, and what happens to real people in the gap.
