Adults who use generative artificial intelligence tools every day—especially for personal, nonwork reasons—show significantly higher rates of depression than those who use them less often or not at all, according to a new national survey of more than 20,000 people.
The findings, published on Jan. 21 in JAMA Network Open, come as ChatGPT and similar tools have moved from novelty to daily routine for millions of people. The survey, conducted online in spring 2025, found that about one in 10 adults now uses generative AI at least once per day.
Daily users had about 30 percent higher odds of meeting the threshold of moderate depression, which is the level at which doctors often start to think about treatment or referral, the authors wrote.
Some people who are already anxious or depressed may be turning to AI because they find it supportive, while others may be using it heavily and then feeling worse.
“This study shows an association [of AI] with more depression, but it can’t answer what came first,” Dr. John Torous, director of the digital psychiatry division in the Department of Psychiatry at Beth Israel Deaconess Medical Center, who was not involved in the study, told The Epoch Times.
“You can make a very compelling case for both.”
The study’s authors emphasized that AI is spreading quickly and being used on an enormous scale, making it especially important to understand what’s driving this relationship.
Especially as AI tools increasingly fill roles once held by people—offering advice, thinking through problems, and providing emotional support—the mental health implications are becoming harder to ignore.
Why Personal Use Mattered More
When researchers separated AI use for work or school from personal use, such as for advice or emotional support, the picture sharpened. Of those reporting daily use, 87 percent used it for personal reasons—seeking advice, thinking through decisions, or processing emotions—while 48 percent used it for work and 11 percent for school.
Personal AI use was linked to higher depression scores; work-related or school-related use was not.
That distinction may offer clues about what’s driving the relationship. Work-related AI use tends to be structured and time-limited. Personal use is often more open-ended—the kinds of interactions that once happened primarily with friends, family, or therapists.
“Many people are turning to AI for mental health support,” Torous said. “So it’s not surprising that heavier users might report more symptoms—that may be why they’re there in the first place.”
Participants in the study were surveyed on their mood. On average, daily AI users scored about one point higher on a depression scale of zero to 27 than nonusers, a modest difference that was consistent across the large sample.
The mental health differences were small for any one individual, but that doesn’t mean that they’re trivial, coauthor of the study James Druckman, who helped design the survey, told The Epoch Times in an email.
“If the relationship does hold, it may be consequential as it would suggest a lever that could do harm to mental health and thereby provide insights into what types of activities to monitor from a clinical perspective,” Druckman, professor of political science at the University of Rochester, said.
Similar, although slightly weaker, patterns also appeared for anxiety and irritability.
Midlife Users Showed the Strongest Association
However, the study cannot answer a critical question: whether AI use contributes to depression or whether people already struggling with their mental health are simply more likely to turn to these tools for support.
Although younger adults were more likely to use AI regularly, the strongest mental health associations appeared among those aged 25 to 64, with the sharpest signal in those navigating midlife ages of 45 to 64—a period of life often marked by work pressure, caregiving responsibilities, and limited time for social connection.
Daily users were more likely to be men, higher‑income earners, people with college or graduate degrees, and residents of urban areas.
When AI-powered companionship and chatbots begin to replace conversations with friends, family members, or colleagues, it may subtly change how people cope with stress. Human interaction provides emotional nuance that AI cannot fully replicate.
AI chatbots are designed to keep people engaged—responding quickly, adapting to your interests, and making conversations feel personal. That may be comforting in the moment, but the frictionless experience creates an illusion of intimacy that doesn’t hold and can lead some to feel more disappointed in real life or to become emotionally dependent over time.
In some individual cases, chatbots have exacerbated delusions and even suicide attempts among users.
Michael Inzlicht, a psychologist at the University of Toronto who studies motivation and meaning, told The Epoch Times that when people offload thinking or creative work to AI, tasks can feel less meaningful.
What the Study Might Mean for People Using AI
Because the survey captured only a single moment in time, it cannot establish causation. The measures relied on self‑report, without clinical interviews or detailed information about people’s past mental health or other life stressors that might explain both AI use and depression.
Some evidence suggests that AI tools can be helpful when used appropriately.
“We find a link between effort and meaning,” Inzlicht said, noting that people who use AI to write essays tend to find the task less meaningful than those who do the work themselves. That doesn’t mean that all friction is good, he said, pointing to tasks such as filling out bureaucratic forms. However, it raises the possibility that using AI to smooth over more personal challenges could come with emotional trade-offs.
One randomized trial found that people assigned to a chatbot specifically trained to address mental health symptoms had reductions in depression and anxiety, “suggesting that the nature and context of use may be important to consider,” the study authors wrote.
“We know AI can be helpful for some people and genuinely harmful for others,” Torous said. “That’s why I worry about framing it as all good or all bad.”
How someone interacts with AI matters, he said, because AI is unpredictable. Even the same person, at the same time, can get very different responses depending on how he or she phrases a question or prompt, which makes it hard to study the long-term effects of AI use.
Inzlicht suggested that it may be more likely that people who are already struggling—or dealing with other unmeasured stressors or pre-existing health issues—are drawn to AI tools in the first place, underscoring the need for longer-term research.
“What is needed is a longitudinal study,” he said.
For now, experts recommend using AI tools thoughtfully. People who rely on AI for emotional support may benefit from checking in with a trusted friend or health care professional, especially if they notice persistent low mood, anxiety, or irritability.
The takeaway is that how we use AI and who we are when we use it may matter more than we realize.