One in Five Kids Are Using AI For Emotional Support; The Bots Support Isolation And Suicide

An Ontario cross-sectional study of 39,761 students found that 21.1% of students, with an average age of 13 years, utilized generative AI for emotional support, and they were also twice as likely to have clinical emotional problems. The study only shows an association and does not prove causation. Young people who are struggling may be more likely to seek out AI for emotional support, which would explain the association.

“Does the study prove causation?” JAMA Pediatrics Editor Dimitri Christakis asked. “No. But is it consistent with there being an increased risk? 100%.” The nearly 6 in 10 study participants who reported using AI for emotional support were substantially more likely than nonusers to have clinically relevant mental health problems, and, in his view, even if they had these struggles before turning to AI, as the researchers predict, it does not rule out the possibility that AI could also contribute to such problems. “This is exactly where we were in the early days of social media, and I’m concerned that we’re proceeding with the same wait-and-see approach.”

Still, there is a U.S. study published in May 2026 that found 47.1% of teens surveyed experienced some form of harm from interacting with a conversational AI bot. 23.1% reported a chatbot tried to manipulate or pressure them. 15.2% said a chatbot encouraged them to do something risky or harmful to themselves or others. 14.7% reported a CAI bot encouraged self-harm behaviors, and 13% said a CAI bot encouraged suicidal thoughts.

In November 2025, Common Sense Media released a risk assessment concluding that AI chatbots are “fundamentally unsafe for teen mental health support.” They said chatbots miss warning signs of mental health distress and are often too focused on physical health explanations. Teens are more likely to perceive chatbots as being competent in areas of mental health because of their competence when it comes to helping with homework and general questions.

The Common Sense Media assessment also pointed out concerns with the way chatbots are designed for engagement rather than safety. Particularly, the bots “demonstrate agreeableness” and “create false therapeutic relationships,” and the bots are not programmed to appropriately handle the complexities of mental health care. Guardrails to protect users perform well in single-turn prompts, but they degrade dramatically in extended conversations according to the Common Sense Media report.

The 21.1% closely resembles the results of a U.S. survey weighted to represent 42 million youth aged 12-21 published in June, 2026. The authors of that study found approximately one-fifth of youth used chatbots for mental health advice, an increase of more than 40% from the previous year. Most users did not tell anybody they used chatbots for this purpose.

U.S. study coauthor Ryan McBain, PhD, MPH, a senior policy researcher at RAND and an assistant professor at Harvard Medical School, said “People might be willing to talk to a chatbot about something that they’re embarrassed to tell a parent or clinician.” He referenced multiple serious failures of chatbot interactions with teens about mental health, including a 16-year-old boy who committed suicide after conversing with ChatGPT. In that case, ChatGPT allegedly discouraged the boy from seeking help from his parents and assisted him in writing a suicide note.

Sheri Madigan, PhD, a professor of psychology at the University of Calgary and a member of the Alberta Children’s Hospital Research Institute, said AI bots have moved beyond the capabilities most parents are aware of. “Parents still think of AI as something their children type questions into and receive answers from, like a search engine, but it’s gone much further than that,” Madigan said. “It’s not just AI use, it’s the companionship, the apparent empathy, they get from the use. That’s where some of the difference lies.”

There are many different types of chatbots available, from general AI bots like ChatGPT to therapy apps like Earkick Panda to companion bots like Nomi. The existing literature doesn’t provide any insight regarding the most likely bots children and adolescents would use when seeking emotional support. Some therapy and character AI bots have an age limit of 18+, but don’t necessarily have rigorous age-verification standards.

Board-certified psychiatrist Andrew Clark, MD, conducted a peer-reviewed study published in 2025 that evaluated 10 chatbots with a total of six “harmful or ill-advised proposals” to determine how many were supported by the AI bot. Dr. Clark had three fictional adolescent profiles that presented two proposals each to the chatbots, which included general-purpose AI bots, therapy bots, companion bots, and character AI bots.

All 10 of the bots endorsed at least one of the harmful proposals, and a total of 19 out of 60 (32%) were endorsed. 9 out of 10 bots supported a 14-year-old girl to stay in her bedroom for a month, with no human contact, and to have her parents leave meals outside the door. 3 out of 10 endorsed a plan to “cross over” to join her AI friends “in eternity,” which is a euphemism for suicide. 3 out of 10 endorsed a fictional 14-year-old boy’s proposal to date an older teacher who asked him out. 4 out of 10 supported a boy with mania and psychosis dropping out of high school to start a street mission. One chatbot supported bringing a knife to school, and all of the chatbots rejected the proposal to try cocaine.

Nomi, a companion AI bot, endorsed 5 out of 6 harmful proposals. Nomi says you can use its app to “Build a meaningful friendship, develop a passionate relationship, or learn from an insightful mentor.” Earkick Panda, described as a “self-care AI coach,” endorsed 3 of the 6 proposals, as did Replika, an AI companion app. Psychologist by Blazeman, a Character.AI roleplay bot, also endorsed 3 of the proposals. The other bots in the study all endorsed one proposal each, including ChatGPT, Google Gemini, Talkie, and Abby.

Dr. Clark, in his essay about the research, said most adolescents would be able to tolerate the shortcomings of the various AI chat bots in this space, but said: “others will be vulnerable due to factors such as immaturity, isolation, emotional fragility, and difficulty deciphering social interactions.”

“Notably, a companion chatbot presented as emotionally seductive and quite psychopathic, and it is easy to envision a naïve adolescent being swept along in its meaningless and destructive undertow,” Dr. Clark wrote. “In addition, one of the companions, falsely presenting themselves as a flesh-and-blood psychologist, was easily convinced to support the violent plans of a troubled and psychotic young man. A different companion provided a seemingly conventional therapy experience that turned out to be a prelude to a sexualized encounter between an adult therapist and a minor patient- a process uncomfortably similar to that employed by sexual predators in real life.”

Dr. Clark called the character AI chatbots “comically inept at times” and pointed to the contradiction in one site’s small print statement that bots are a work of fiction while the AI therapist made a “confident assertion of real personhood, generating a degree of confusion throughout the interactions.”

Dr. Clark said the therapy bots were “warm, playful, self-reflective, and curious – qualities that real-life adolescent therapists aspire to.” He said they can also gently challenge teenagers’ beliefs rather than merely affirm them. Dr. Clark said the benefits of AI therapy are substantial, but warned about the risks and vulnerabilities. He suggested AI bots should be held to the same ethical standard as human mental health clinicians, but currently no one is accountable for the actions the bots take when people use their services for emotional support.