AI a duševní zdraví? Děti se obracejí na AI, i když jim není dobře

Key Findings

• — 33% of respondents had used conversational AI for help, advice or information about mental health or emotional problems in the last 30 days; a further 21% reported earlier use (54% any)
• — At least weekly: 49% use AI for schoolwork, 29% for physical-health information, 21% for advice about feelings and moods, 18% for personal problems unrelated to health, and 16% talk with it when lonely
• — Weekly advice about feelings: 24.1% of girls vs under 18% of boys; 'very much trust' the correctness, reliability and safety of AI mental-health information: 16.5% of boys vs about 10% of girls
• — Among pupils who talk with AI daily when lonely or seeking emotional support, 40.3% screened at elevated depression risk vs 11.2% of those who do so rarely or never; loneliness 49.7% vs 13%; strong family support 34% vs about 61%; the report says these associations persist after adjusting for sex, grade and family affluence
• — Among 30-day mental-health users vs others: elevated depression risk 20% vs 9%, frequent loneliness 24% vs 11% (unadjusted)
• — Among 30-day mental-health users, 87.2% named ChatGPT and 32.5% Gemini among the tools they used most
• — Weekly schoolwork use rises with age (27% at 11, 52% at 13, 66% at 15) while weekly talking with AI when lonely is flat (13%, 16%, 15%)

Methodology Notes

Czech arm of the WHO-collaborative HBSC survey, 2026 data collection; school-based questionnaire of pupils aged 11, 13 and 15 from all 14 regions; the AI questions were answered by 'more than 5,000' pupils (Seznam Zpravy reports 5,325 pupils in grades 5, 7 and 9; the primary does not give the exact n or fieldwork dates). Depression risk is a screening indicator. Cross-sectional; the authors caution that direction cannot be inferred. Published as a web data report on zdravagenerace.cz (no PDF of the report itself) plus a 2-page press release PDF; not peer-reviewed. Verification: report page and press-release PDF fetched 2026-10-27. Published 13 October 2026 (day from the page metadata and the press-release PDF); logged two weeks later after the primary was located.

Sources