Daily AI Use Was Associated With More Depressive Symptoms in a US Survey

A US survey found that adults who reported using generative AI every day also reported slightly more depressive symptoms on average. Daily users had about 29% higher odds of screening positive for at least moderate depressive symptoms than nonusers after the researchers adjusted for measured demographic factors.

The finding is an association, not evidence that ChatGPT, Gemini, Copilot, or another tool causes depression. The study measured AI use and symptoms at the same time, so it cannot determine which came first. People experiencing loneliness, anxiety, depression, work stress, or other difficulties may also be more likely to turn to AI frequently.

What the JAMA Network Open study measured

The study in JAMA Network Open analysed responses from 20,847 adults in all 50 US states and Washington, DC. Participants joined a nonprobability internet survey conducted from April 20 to May 27, 2025.

• 2,152 participants, or 10.3%, said they used generative AI at least daily.
• 1,053 used it every day, and 1,099 used it multiple times per day.
• Participants completed the PHQ-9 questionnaire for depressive symptoms; the researchers also examined brief measures of anxiety and irritability.
• Daily or more frequent use was associated with an adjusted odds ratio of 1.29 for at least moderate depressive symptoms compared with nonuse.

An odds ratio of 1.29 is often shortened in headlines to “29% higher odds.” It is not the same as saying that 29% of users became depressed or that their absolute risk rose by 29 percentage points. The questionnaire is a screening measure, not a clinical diagnosis by itself.

The association differed by age and purpose

The adjusted association was most apparent among adults aged 25–44 and 45–64. For the 45–64 group, the reported odds of at least moderate depressive symptoms were about 54% higher among daily users than nonusers. Personal AI use showed a stronger association than work or school use.

Those subgroup results can help researchers decide what to study next, but they do not establish that an age group is harmed by AI. The more comparisons a study makes, the more carefully individual subgroup findings need to be interpreted and replicated.

Why the study cannot establish cause

This was a cross-sectional, self-reported survey. It has several limits:

• No time order: AI use and mood symptoms were recorded in the same survey wave.
• Reverse causation: existing symptoms could contribute to heavier AI use.
• Unmeasured factors: sleep, isolation, health, job conditions, personality, or life events could affect both AI use and mood.
• Broad exposure: “generative AI use” combines different products, purposes, session lengths, and conversation styles.
• Survey selection: participants opted into an internet panel, even though the researchers weighted the results to better reflect the US population.

These limits make “daily AI use causes depression” an unsupported conclusion. The study's authors described the effect as modest and called for research that can test direction and mechanisms.

Newer follow-up evidence did not find a subsequent increase

A separate 2026 study used several survey waves to emulate a randomized trial. Among the 3,109 participants who returned for follow-up, frequent AI use at baseline was not significantly associated with a later change in PHQ-9 scores. The authors concluded that their result did not support the idea that AI use causes greater depressive symptoms for most US adults, while noting that harm in vulnerable individuals could not be excluded.

This second study is also observational and had substantial loss to follow-up, so it does not settle the question. Together, the studies show why a single association should prompt better research rather than a causal headline.

General chatbots and therapeutic chatbots are different

A general assistant used for work, companionship, or open-ended advice is not the same intervention as a chatbot built around a defined psychological program and tested in clinical trials. A 2026 meta-analysis of 29 randomized trials found that cognitive-behavioural-therapy-oriented chatbots were associated with short-term reductions in depressive and anxiety symptoms. However, the authors rated the overall certainty of the evidence from very low to low because of bias and substantial differences between studies.

That evidence does not make a consumer chatbot a therapist. A language model can produce confident, empathic-sounding text without knowing a person's full history, assessing nonverbal cues, accepting professional responsibility, or reliably handling a crisis. TipsMake's review of ethical and safety risks in AI counselling explains why prompts alone do not create clinical oversight.

How to use AI without treating it as mental health care

• Use it for low-stakes tasks such as organising thoughts, drafting questions for an appointment, or finding terms to discuss with a clinician.
• Check health claims against reputable sources and do not let a chatbot diagnose, prescribe, or change treatment.
• Notice whether AI sessions are replacing sleep, daily responsibilities, in-person relationships, or professional care.
• Set a purpose and time limit before an emotionally charged conversation.
• Avoid sharing identifying health information unless you understand the service's privacy and retention settings.

TipsMake's principles for using AI companions provide a practical boundary-setting checklist, while the guide to situations where ChatGPT should not make the decision covers health and other high-stakes limits.

When to seek human help

Depression involves more than having a bad day. The World Health Organization describes persistent low mood or loss of interest or pleasure, often with changes in sleep, appetite, energy, concentration, self-worth, or thoughts about death. Effective treatments are available.

If symptoms persist, interfere with everyday life, or feel difficult to manage, contact a doctor or licensed mental health professional. If you may act on thoughts of self-harm or suicide, contact local emergency services or a crisis service in your country now and reach a trusted person who can stay with you. A chatbot should not be the only source of support in an urgent situation.