Generative AI Use and Mind Health Outcomes

Key Findings

• — 66.7% of 264,085 respondents reported using generative AI chatbots; 89.0% of 18-34-year-olds were active users versus 55.4% of those 35 and older, and 41.3% of 18-34 users used it several times a day or constantly versus 18.2% of users aged 35+.
• — Mean MHQ was higher at several-days-a-week use than at rare use, then fell sharply at daily and constant use; the drop from peak to constant use was 25.1 MHQ points at ages 18-24, 17.4 at 25-34 and 7.5 at 65+.
• — Focus and concentration declined and addictions and cravings rose at every higher level of use; constant use was associated with worse scores on all 47 MHQ items. Hallucinations and suicidal thoughts were among the problems that fell most at moderate use before rising again at high use.
• — Among 18-34-year-old users, 36.7% used AI for both emotional and cognitive purposes; the share reporting emotional use (companionship, emotional support or counselling, romantic or intimate conversation) rose from 20% of rare users to 55% of constant users.
• — Users with emotional plus cognitive use had MHQ scores 26-32 points lower than cognitive-only users at every frequency of use; the inverted-U pattern held in both groups.
• — The authors state the data are cross-sectional and cannot establish direction; they favour a reciprocal interpretation and report that sleep, diet, exercise and trauma prevalence did not differ among daily users (data not shown).

Methodology Notes

Cross-sectional, non-probability online survey (Global Mind Project MHQ assessment: 47 items spanning 27 capacities and 20 problems on a 9-point life-impact scale, aggregate -100 to +200). AI items added 2025-06-16; data to 2026-04-12; 264,085 adults 18+, of whom 176,198 GenAI users were analysed. Frequency asked on five levels (rarely to constantly); purposes: ten cognitive and three emotional. Analysis reports group means by age band, sex, frequency and use type, with significance and effect sizes; no confidence intervals or cell sizes are printed. Not peer reviewed: the foreword (signed by Chief Scientist Tara Thiagarajan; no author list) says an academic publication will follow and that the data are available to academic researchers. A section heading says emotional use carries a burden 'dominated by symptoms of psychosis', but the report gives no item-level psychosis figures for emotional users. Date: the Sapien Labs research page lists the report as 'September 30, 2026'; the PDF cover reads 'August, 2026'; the PDF's Last-Modified header is 2026-09-29 22:22 GMT.

Sources