- A new study investigates associations between the use of artificial intelligence (AI) and depressive symptoms.
- They conclude that those who used AI most often were more likely to experience depressive symptoms than those who never used it.
- Interestingly, their results varied significantly between age groups.
A group of researchers published a paper in
Analyzing questionnaire data from almost 21,000 participants in 50 states, the researchers concluded that high AI usage was linked to a greater likelihood of experiencing symptoms of depression, anxiety, and irritability.
Although AI only entered the scene a handful of years back, it is already a constant feature of many people’s lives. Whether asking ChatGPT how to cook a chicken curry or using it to summarise data tables at work, it is impossible to ignore.
As with the advent of any new technology, understanding its impact on users’ health is an important area of research.
Already, there is some anecdotal evidence that chatbots can feed people’s delusions and even encourage suicidal thoughts. Despite these worrying early signals, little academic research has focused on AI use and mental health. The latest study goes some way toward filling that gap.
The researchers accessed survey data that was collected in April and May of 2025. This included 20,847 participants across 50 states, all of whom were 18 years or older. This data was well-balanced, containing information from a range of age groups, ethnicities, and genders in each state.
As part of the survey, all participants were asked, “How often do you use the following technologies or products? — Artificial intelligence (AI).”
The available responses were never, once or twice, about once a month, about once a week, multiple times a week, every day, and multiple times a day. They were also asked whether it was for personal use, work, or school.
The questionnaire contained questions about mental health using the 9-item Patient Health Questionnaire (PHQ-9), which collects information on depressive symptoms. It also included a 2-item Generalized Anxiety Disorder scale (GAD-2), which derives from the 7-item version, and the 5-item Brief Irritability Test (BITe).
Alongside this information, the researchers also had access to personal data, such as household income and educational achievement.
The analysis showed that 10.3% of the sample used AI every day, which includes 5.3% who use it multiple times each day.
Among daily users, around half used it for work, 11.4% for school, and 87.1% for personal reasons.
Unsurprisingly, AI usage varied by demographic. Those who used AI the most were more likely to be younger, male, have a higher level of education, have a higher household income, and live in an urban rather than rural setting.
Medical News Today reached out to the corresponding author, Roy H. Perlis, MD, MSc, professor of psychiatry at Harvard Medical School and clinical investigator at Mass General Research Institute, both in Boston, MA.
“Odds of at least moderate depression (the threshold at which clinicians would typically refer people for evaluation and treatment) were 30% greater among those with at least daily AI use,” he explained. When looking at anxiety and irritability, a similar pattern emerged.
When they dug a little deeper into the data to look at whether the type of usage mattered, they found that this link to mental Health problems was only significant for personal, rather than work or school use.
The researchers found no differential associations between AI usage and gender, but there was an association with age group. Specifically, AI usage was associated with more depressive symptoms in the 25 to 44 and 45 to 64 age groups.
As with any research, especially in a new field, each new study generates more questions than answers.
First and foremost, the question of cause and effect. “We don’t know if it is cause, effect, or neither,” Perlis told MNT. “Only a randomized trial could tell us for sure, although following people over time might help.”
On this question, he also told us that it is “certainly possible that people with a greater extent of depressive symptoms use AI more often.”
Another question left open is that of the relationship with age. People younger than 25 or older than 64 did not show the same association between depressive symptoms and AI usage. “This group may be more likely to use AI in particular ways or particular settings, but we do not yet know,” Perlis told MNT.
It may be that certain, specific forms of AI usage might be more problematic than others. The survey used in this study was not designed to capture that information, but Perlis suspects this might be an important area of future research:
“As a psychiatrist, I certainly worry about people using AI as a substitute for social interaction. In fact, this was my motivation for looking at this association in the first place.”
MNT also contacted John Puls, LCSW, MCAP, a psychotherapist and addiction specialist at Full Life Comprehensive Care, who was not involved in the study.
“The results do not surprise me at all,” he told us. “The use of large language AI will often prevent you from engaging in meaningful conversations with others and getting support from them.”
This, he explained, could lead to increased isolation and loneliness. Puls also said that “when AI does everything, including your thinking and problem solving for you, it is natural to start feeling devoid of purpose, which will cause depression.”
Although the results of this study might be perceived as worrying, Perlis hopes that the personal use of AI could still have benefits if deployed in the right way.
“There is no question that chatbots may be promising interventions for people who otherwise cannot access talk therapy,” he explained. “But only if those chatbots are carefully designed and monitored, and ideally used in conjunction with a human therapist.”
Importantly, he also told us that the current study does not exclude the possibility that some individuals benefit from AI.
“Indeed, AI can be a force for good,” Puls to MNT. “Reducing someone’s work burden and improving mental health. But if that is done at the expense of real human connection, it will have significant negative consequences.”
The large size and scope of the questionnaire allow us to have a certain level of confidence in the results. However, as Perlis told MNT, “as an internet survey, our sample is not perfectly representative, and there may well be biases.”
MNT also reached out to Owen Muir, MD, DFAACAP, a psychiatrist at Radial, who echoed this sentiment. “If you wonder who would be more likely to talk to a Chatbot multiple times a day, that is selecting for people who are lonely, and more likely to be depressed.”
“To increase confidence in our results,” Perlis told us, “I hope other groups will try to replicate these results using other samples. I would also be interested in seeing longitudinal studies where we see how mood changes over time with AI use.”
“And of course, the most direct test would be a randomized trial, though these sorts of studies are difficult to do.”
