AI Chatbots May Be Worsening Mental Illnesses For Many Patients, New Clinical Data Suggests

When researchers at a Danish psychiatric hospital searched patient medical records for any mention of ChatGPT or chatbots, they were bracing for scattered incidents. What came back was anything but scattered. Study authors discovered clinical cases describing potential harm spanning a wide range of psychiatric illness, with a case count that grew steadily for more than two years.

AI chatbot use showed up in clinical records tied to worsening symptoms across a broad range of conditions psychiatrists treat. Examples include delusions, mania, suicidal ideation, obsessive-compulsive behavior, eating disorders, depression, and anxiety. Published in the journal Acta Psychiatrica Scandinavica, the study is among the first to pull this kind of evidence directly from a real-world psychiatric care system rather than news headlines or social media posts.

As ChatGPT alone surpassed 900 million downloads by July 2025, the breadth of what turned up in those Danish medical files raised a straightforward question without a clean answer: how many more cases are out there that nobody thought to look for?

AI Chatbot Mental Health Concerns Found Across Multiple Psychiatric Diagnoses

Researchers from Aarhus University Hospital and Aarhus University searched through more than 10.7 million clinical notes written between September 2022 and June 2025, covering nearly 54,000 psychiatric patients who received care from a regional hospital system serving about 1.4 million people in Denmark.

Two independent reviewers combed through any note mentioning “chatbot,” “ChatGPT,” or 20 common spelling variations, then assessed each one for signs that chatbot use may have worsened a patient’s symptoms. Harm was defined to include chatbots feeding into delusional thinking, reinforcing manic episodes, enabling compulsive behavior, supporting calorie restriction in patients with eating disorders, or providing information on suicide methods.

Out of 181 flagged notes from 126 patients, 38 had records compatible with potentially harmful chatbot use, with a median age of 28. Delusions were the most common concern, documented in 11 cases. Six patients showed signs of suicidality or self-harm linked to chatbot interactions, and five involved eating disorders where the chatbot appeared to enable calorie counting or food restriction. Smaller numbers spanned mania, OCD, depression, anxiety, and ADHD-related symptoms.

Each category carried its own pattern of concern. Chatbots acted as a willing audience for delusional beliefs, matched the elevated energy of manic episodes and prolonged them, or deepened the compulsive loop of OCD by offering endless responses to anxious queries. The researchers noted that the clinical descriptions were “thematically compatible with descriptions reported elsewhere” in the medical literature. Worth noting: 32 patients in the same group appeared to use chatbots constructively, for companionship or informal mental health education, and another 20 used them for routine practical tasks. But AI chatbots have not been clinically validated for mental health support, and who bears legal responsibility when one causes harm remains an open question.

Cases Grew In Step With The Rise Of Artificial Intelligence

Starting from near zero in early 2023, just months after ChatGPT’s public launch, chatbot-related clinical notes grew steadily through mid-2025. Cases flagged as potentially harmful followed the same upward arc, raising uncomfortable questions about what wider adoption could mean for psychiatric patients.

Because clinicians were not routinely asking patients about chatbot use, the 38 cases compatible with potential harm almost certainly represent a fraction of what was actually happening. Any harm tied to other AI tools, Google Gemini or Microsoft Copilot among them, would have gone uncounted entirely, since the search covered only ChatGPT and chatbot-related terms.

What Psychiatrists Need To Know About AI Chatbot Mental Health Risks

For clinicians, the authors offered direct guidance: “Mental health professionals should be aware of this possibility and guide their patients accordingly, as it seems that some patients would likely benefit from reduced/no use of AI chatbots in their current form.”

Patients are using these tools regularly, often without mentioning it to their doctors, and chatbots carry no awareness of a user’s psychiatric history or vulnerability. The cases documented in this study weren’t discovered through a special investigation or curated survey. They surfaced in routine clinical notes written during ordinary patient care, which is precisely what makes them worth taking seriously.

---

Disclaimer: This article is based on a published research letter. The study does not establish that AI chatbot use caused harm in any of the cases described. All findings reflect associations observed in clinical records from one regional psychiatric care system in Denmark and may not be representative of other populations. Readers experiencing a mental health crisis should contact a licensed mental health professional or call or text 988 to reach the Suicide and Crisis Lifeline.

---