Daily use of generative artificial intelligence (AI) was linked to increased risk for depressive symptoms and other adverse outcomes, particularly in younger and middle-aged adults, new research showed.

In a US survey study with nearly 21,000 adult participants, the risk for at least moderate depression was 30% higher in daily AI users than in nonusers. The association was particularly strong among people aged 25-64 years, who were 30-50% more likely to report depressive symptoms.

“This emerging picture is consistent with previous observations regarding social media use,” lead author Roy H. Perlis, MD, Massachusetts General Hospital, Boston, and colleagues wrote.

The findings were published online on January 21 in JAMA Network Open.

Deeper Dive

Use of generative AI increased sharply following the release of ChatGPT 3.5 (OpenAI) in late 2022, investigators noted. Some studies suggest excessive daily use may be associated with dependence and social isolation. And anecdotal reports point to possible harms to mental health, including delusions in vulnerable individuals, they added.

To investigate further, researchers used data from a 50-state internet survey conducted for 1 month in 2025, with 20,847 adult participants (mean age, 47 years; 50% men). Of these individuals, 10.3% reported at least daily use of AI, and 5.3% reported using it multiple times per day.

Among those with daily or greater AI use, the biggest reason given for the use was for personal reasons (87.1%), followed by work (48%) and school (11.4%).

Outcome measures included the nine-item Patient Health Questionnaire (PHQ-9) to measure depressive symptom severity, the two-item Generalized Anxiety Disorder (GAD-2) screen, and the Brief Irritability Test.

Men, younger adults, and those with higher education were among those who most commonly used AI at least daily.

In adjusted models, greater use of AI was linked to greater PHQ-9 scores (both daily use and multiple times per day, P < .001), greater GAD-2 scores (P < .001 for both), and greater irritability scores (P = .001 and .004, respectively).

Compared to nonuse, daily use was also linked to greater odds of having moderate or severe depressive symptoms in the full population (odds ratio [OR], 1.29), as well as in those aged 25-44 years (OR, 1.32) and 45-64 years (OR, 1.54) but not in those aged 65 years or older.

“This survey study found that generative AI use was associated with modestly but statistically significantly greater depressive and other negative affective symptoms, warranting efforts to understand the potential for a causal relationship,” the investigators wrote. 

The findings also suggest that “there is a need to understand the nature and mechanism of this association and its heterogeneity across age-defined subgroups,” they added. 

Several Caveats

In a statement from the UK nonprofit Science Media Centre (SMC), David Harley, PhD, co-chair of the British Psychological Society’s Cyberpsychology Section, said that the findings seem to replicate those of previous studies tying depression and anxiety to digital engagement.

However, Harley, who was not involved with the current research, noted several caveats with the findings, including whether they could be generalizable to other countries.

“Is American use of AI the same as the UK and other countries (how common is it in the workplace, for instance)? Also, we know that the context of use is very important if we really want to understand AI’s relationship to mental health,” he said.

Also commenting for SMC, David A. Ellis, PhD, chair of Behavioral Science at the University of Bath, Bath, England, cautioned that use of AI relied on self-reports, calling that a “far from perfect” measure of technology use.

“Assuming the effect is not statistical noise, what is suggested is that AI use may harm mental health. However, there are three equally plausible explanations for the direction of the relationship reported in this work,” Ellis added.

The first two are that AI use may lead to depression or that depression may lead to AI use, he noted. 

“Or there is in fact a third variable that has not been measured (eg, social isolation). Indeed, the direction could be entirely reversed from what the title of the paper implies,” Ellis concluded.

The survey was funded by the National Science Foundation and the National Institutes of Health. Perlis reported receiving personal fees from Genomind, Alkermes, and Circular Genomics, outside the submitted work. Two other investigators reported receiving grants from the two survey funders, outside the submitted work. The other researchers and Harley reported no relevant financial relationships. Ellis reported current or past financial relationships with UK Research and Innovation, the National Institute for Health and Care Research, We Adapt, Duco, Plexal, the BBC, Girton College, and the University of Cambridge. He is also a member of a Department for Science, Innovation and Technology research commission.