What French Teenagers' Experiences Tell Us About the Future of Online Child Safety

Key Findings

• — Among the surveyed 11-17-year-olds, 78% use general-purpose generative AI, 55% voice assistants, 50% AI integrated into social media and 32% AI companions
• — 56% of 11-12-year-olds use at least one type of generative AI at least once a week; regular AI-companion use is reported by 16% of 11-12-year-olds, 13% of 13-15-year-olds and 15% of 16-17-year-olds, although most services' terms exclude these ages
• — Parents underestimate younger children's use: 40% of 11-12-year-olds report regular general-purpose AI use, while parents estimate 26%
• — Among teenagers who use generative AI (N=764), 92% say they spend more time than they would like on at least one tool: 59% for AI companions, 57% for AI in social media, 45% for general-purpose AI and 43% for voice assistants
• — Confiding is a more common reason for use among the youngest: 14% of 11-12-year-old users use AI to confide in it, against 6% of 13-15-year-old users (9% of all teen users)
• — 38% of all surveyed teenagers report experiencing negative emotions at least weekly, rising to 44% of regular generative-AI users and 50% of regular AI-companion users (N=136); the authors state this is not evidence that AI causes distress
• — 44% of parents whose children use generative AI say they do not feel at ease when their child uses it; 42% of parents do not feel supported at all in helping their children navigate it

Methodology Notes

YouGov online panel survey in France, 24-30 June 2026: 1,138 parents of children aged 11-17 (representative by age, gender, region and socio-professional category) and 970 teenagers aged 11-17 with parental consent (sized for analysis by age band 11-12, 13-15, 16-17; not described as representative). Respondents were given definitions and examples for each AI category (e.g. ChatGPT; Character.AI; My AI on Snapchat; Siri). A self-reported 'vulnerability score' built from five well-being items (parents: three) is used for sub-group analysis and is not a clinical measure. Cross-sectional; several sub-groups are small (regular AI-companion users N=136). Note the YouGov PDF's embedded metadata title is a stale template ('Brand tracker Q4 2023 France, Germany, Italy'); the content is the interface study.

Sources