Daily AI Use Linked to More Depressive Symptoms in Study of 20,847 US Adults

Adults who used generative artificial intelligence (AI) every day had 29% higher adjusted odds of reporting at least moderate depressive symptoms than less frequent users, according to a study of 20,847 Americans published in JAMA Network Open.

The study, led by Roy H. Perlis, MD, MSc, of Massachusetts General Hospital, found that 32.0% of daily AI users met a commonly used screening threshold for moderate depressive symptoms, compared with 25.9% of those using AI less frequently.

The researchers also identified associations with anxiety and irritability, with particularly notable depression-related findings among adults ages 25–64.

The results do not establish that AI causes depression. Because the researchers measured AI use and mental health at the same time, they could not determine which came first.

A visibly stressed woman with tired eyes rests her head on one hand while looking at a laptop in a bright, comfortable home.

A study of 20,847 US adults found that daily generative AI users reported more depressive symptoms than less frequent users, although the research did not establish causation.

Key Takeaways

  • 20,847 US adults participated in the study, including 2,152 (10.3%) who used generative AI at least daily.
  • 32.0% of daily AI users screened positive for at least moderate depressive symptoms, compared with 25.9% of less frequent users.
  • Daily AI use was associated with 29% higher adjusted odds of meeting the depression screening threshold (OR 1.29; 95% CI, 1.15–1.46).
  • Adults ages 25–64 showed particularly notable associations between AI use and depressive symptoms.
  • Personal AI use was associated with slightly higher depression scores.
  • Anxiety and irritability scores were also higher among daily AI users.
  • Causation remains unknown. People experiencing depressive symptoms may use AI more frequently, or other factors may contribute to both.

How Researchers Studied AI Use and Depression

Researchers analyzed data from the Civic Health and Institutions Project, a US internet survey conducted between April 20 and May 27, 2025.

The study included adults from all 50 states and Washington, DC, with an average age of 47.3 years.

Participants reported how frequently they used generative AI and whether their use was personal, professional, or educational.

Mental health was assessed using three established instruments:

  • PHQ-9: A nine-question depression screening tool scored from 0 to 27.
  • GAD-2: A two-question anxiety screening tool.
  • BITe-5: A five-question measure of irritability.

A PHQ-9 score of 10 or higher is commonly used to identify at least moderate depressive symptoms. It is not, by itself, a clinical diagnosis of depression.

Researchers adjusted their analyses for demographic characteristics, including age, gender, education, income, race, ethnicity, and urban or rural residence.

The study used a nonprobability online sample with statistical weighting to improve demographic representation.

Daily AI Users Had Higher Depression Screening Scores

Of the 20,847 participants, 2,152 reported using AI at least daily.

The remaining 18,695 used AI less frequently, including those who never used it.

The researchers found the following differences:

Mental health measure Daily AI users Less frequent users
Participants 2,152 18,695
Average PHQ-9 depression score 7.4 6.3
PHQ-9 score of 10 or higher 32.0% 25.9%
Average GAD-2 anxiety score 1.8 1.6
Average BITe-5 irritability score 13.7 12.8

Daily AI users were 6.1 percentage points more likely to screen positive for at least moderate depressive symptoms.

After adjusting for demographic differences, daily use was associated with 29% higher odds of meeting the screening threshold (OR 1.29; 95% CI, 1.15–1.46).

This does not mean daily AI use increases someone’s risk of developing depression by 29%. An odds ratio describes an association between groups and cannot establish what caused the difference.

How Significant Was the Difference?

Daily AI users scored approximately 1.1 points higher on the PHQ-9 than less frequent users.

Compared with participants who never used AI, the adjusted regression coefficients for depressive symptoms were:

  • Once-daily users: β = 1.08 (95% CI, 0.55–1.62).
  • Multiple-times-daily users: β = 0.86 (95% CI, 0.35–1.37).

Both associations were statistically significant.

However, differences of approximately one point are modest on the PHQ-9’s 27-point scale. Statistical significance does not necessarily indicate a clinically meaningful difference for an individual.

The findings also did not show a consistent dose-response pattern. Multiple-times-daily use was not associated with a larger coefficient than once-daily use.

Which Age Groups Had the Strongest Associations?

Age appeared to influence the relationship between generative AI use and depressive symptoms.

The researchers reported higher depression score associations among two age groups:

Age group Adjusted PHQ-9 coefficient 95% confidence interval
25–44 years 1.22 0.70–1.74
45–64 years 1.38 0.72–2.05

Adults ages 45–64 showed the larger estimated association with depression symptom scores.

These results suggest that the relationship between AI use and depressive symptoms may differ across age groups.

However, the researchers did not establish why these differences occurred. The findings should not be interpreted as evidence that adults in either age group are inherently more vulnerable to AI.

Was Personal AI Use More Closely Associated With Depression?

Among daily AI users, personal activities were the most commonly reported purpose.

Purpose of AI use Percentage of daily users
Personal use 87.1%
Work 48.0%
School 11.4%

Participants could report more than one purpose.

After accounting for demographic characteristics and the different purposes of AI use, participants reporting personal AI use had PHQ-9 scores averaging 0.31 points higher than those who did not report personal use (95% CI, 0.10–0.52).

This was a statistically significant but small association.

The survey did not examine what people discussed with AI systems. Personal use could include entertainment, writing, planning, information seeking, emotional support, or other activities.

Consequently, the results cannot establish whether emotionally significant conversations with chatbots contributed to the observed association.

What Have Other Studies Found About AI and Depression?

Two earlier studies provide useful comparisons. One found an association between chatbot use and depressive symptoms among university students, while another found improvements in mental health symptoms following a specialized AI intervention.

Chinese University Study Found Higher Depression Scores Among Chatbot Users

A 2025 study published in the Journal of Affective Disorders examined AI chatbot use among 1,004 Chinese university students.

Researchers led by Xing Zhang reported that:

  • 45.8% of students had used AI chatbots during the previous month.
  • Among users, 78.5% used chatbots one to three days per week.
  • Chatbot users reported significantly higher depressive symptom levels than nonusers.
  • Chatbot use and dependence were positively associated with depression scores among users, with coefficients ranging from 0.14 to 0.20.

The researchers found no direct statistically significant association between chatbot-related behaviors and their separate measures of overall mental well-being or resilience.

Like the US study, this research was cross-sectional and relied on self-reported behavior.

The findings suggest that associations between AI chatbot use and depressive symptoms are not limited to American adults. However, the two studies cannot determine whether AI use preceded the reported symptoms.

Randomized Therabot Trial Found Improvements in Mental Health Symptoms

In contrast, a randomized controlled trial published in NEJM AI found that a specialized generative AI chatbot improved certain mental health symptoms.

Researchers led by Michael V. Heinz evaluated Therabot, an AI system specifically developed for mental health treatment.

The trial enrolled 210 adults experiencing clinically significant depression or anxiety symptoms, or elevated risk for feeding and eating disorders.

Participants were randomly assigned to receive four weeks of Therabot access or to a waitlist control group.

Among participants with depressive symptoms, average improvements were:

Time point Therabot group Control group
4 weeks 6.13-point reduction 2.63-point reduction
8 weeks 7.93-point reduction 4.22-point reduction

The Therabot group experienced greater improvements in depression scores, with benefits also reported for anxiety and eating disorder-related outcomes.

However, Therabot was a purpose-built mental health intervention tested under controlled research conditions.

Its results cannot be generalized to general-purpose chatbots or unrestricted AI conversations. The trial was also relatively small, and larger studies are needed to establish long-term safety and effectiveness.

Why Do the Studies Show Different Results?

The three studies examined different populations and types of AI interaction.

Study Participants Study design Main finding
Perlis et al., 2026 20,847 US adults Cross-sectional survey Frequent AI use associated with higher depressive symptoms
Zhang et al., 2025 1,004 Chinese students Cross-sectional survey Chatbot use and dependence associated with higher depression scores
Heinz et al., 2025 210 US adults Randomized controlled trial Specialized therapeutic AI improved mental health symptoms

The observational studies measured associations between reported AI behavior and existing symptoms.

The Therabot trial tested whether a specific intervention could improve symptoms over time.

These findings are not necessarily contradictory. The effects of AI may depend on how a system is designed, what it is used for, and the circumstances of the person interacting with it.

The available evidence does not support treating every type of AI use as psychologically equivalent.

Does AI Use Cause Depression?

There is currently insufficient evidence to conclude that everyday generative AI use causes depression.

The US study identified an association, but several explanations remain possible.

People experiencing depression may use AI more frequently for information, distraction, companionship, or reassurance.

Alternatively, certain AI interactions could influence psychological well-being. For example, prolonged use might affect sleep or the time available for other activities.

A third possibility is that underlying factors, including stress, loneliness, existing psychiatric conditions, or demanding work schedules, contribute to both frequent AI use and depressive symptoms.

These explanations remain hypotheses. The study did not establish which, if any, accounts for its findings.

Longitudinal research and randomized trials examining specific forms of AI use are needed to clarify the relationship.

What Were the Study’s Main Limitations?

The researchers acknowledged several limitations affecting interpretation.

Cross-sectional design: AI use and depressive symptoms were measured at the same time, preventing conclusions about cause and effect.

Self-reported behavior: Participants estimated their AI use rather than providing verified usage records.

Unknown duration and content: Researchers did not establish how much time participants spent interacting with AI or what their conversations involved.

Possible confounding: Demographic adjustment cannot eliminate the influence of unmeasured factors, including preexisting psychiatric conditions.

Online recruitment: The nonprobability survey sample may not fully represent all US adults despite statistical weighting.

Screening rather than diagnosis: The PHQ-9 identifies depressive symptom severity but does not independently diagnose major depressive disorder.

These limitations are particularly important when interpreting the findings as evidence of potential harm.

Should You Be Concerned About Using AI Every Day?

The study does not establish that daily AI use is harmful or that people should stop using it.

For individuals who regularly interact with AI, the more practical consideration is whether those interactions support or interfere with everyday functioning.

Consider whether AI use is affecting:

  • Sleep and daily routines.
  • Relationships and opportunities for social interaction.
  • Work, education, or other responsibilities.
  • Emotional well-being, including reliance on repeated reassurance.

These are general considerations, not behaviors identified as causes of depression in the study.

AI can help users gather information and organize thoughts. However, general-purpose chatbots should not replace professional mental health assessment or treatment.

Persistent sadness, loss of interest in usual activities, or symptoms affecting everyday functioning warrant evaluation by a qualified healthcare professional.

Researchers have not established a safe or harmful amount of daily AI use, and the available evidence does not justify recommending a specific usage limit.

Funding and Disclosures

The US survey received partial funding from the National Science Foundation and National Institutes of Health. The authors reported that funders had no role in the study’s design, analysis, interpretation, or publication decisions.

Lead author Roy H. Perlis disclosed personal fees from Genomind, Alkermes, and Circular Genomics outside the submitted work. Other authors reported research grants outside the submitted work.

Perlis serves as editor in chief of JAMA+ AI and an associate editor of JAMA Network Open. Coauthor Faith M. Gunning is also an associate editor of JAMA Network Open. Neither participated in editorial decisions concerning the manuscript.

The authors disclosed using Anthropic Claude Sonnet 4 to assist with debugging analytical and figure-generation code, with Perlis accepting responsibility for the code.

The article was published January 21, 2026, and corrected February 11, 2026, to fix a typographical error in coauthor Ata A. Uslu’s name. The correction did not affect the study findings.

Final Thoughts

A study of 20,847 US adults found that daily generative AI use was associated with higher depressive symptom scores and 29% higher adjusted odds of screening positive for at least moderate depressive symptoms.

The association was particularly notable among adults ages 25–64, while personal AI use also showed a small relationship with higher depression scores.

Earlier research among Chinese university students identified a similar association. In contrast, a randomized trial found that a specialized AI mental health chatbot improved depressive symptoms.

The evidence suggests that the relationship between AI and mental health may depend on the type and purpose of use. Whether routine generative AI use contributes to depression remains unknown.

References

Heinz, M. V., Mackin, D. M., Trudeau, B. M., Bhattacharya, S., Wang, Y., Banta, H. A., Jewett, A. D., Salzhauer, A. J., Griffin, T. Z., & Jacobson, N. C. (2025). Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI, 2(4), AIoa2400802. https://doi.org/10.1056/AIoa2400802

Perlis, R. H., Gunning, F. M., Uslu, A. A., Santillana, M., Baum, M. A., Druckman, J. N., Ognyanova, K., & Lazer, D. (2026). Generative AI use and depressive symptoms among US adults. JAMA Network Open, 9(1), e2554820. https://doi.org/10.1001/jamanetworkopen.2025.54820

Zhang, X., Li, Z., Zhang, M., Yin, M., Yang, Z., Gao, D., & Li, H. (2025). Exploring artificial intelligence (AI) chatbot usage behaviors and their association with mental health outcomes in Chinese university students. Journal of Affective Disorders, 380, 394–400. https://doi.org/10.1016/j.jad.2025.03.141