Same question, different answers: the gendered impacts of AI on young people

Key Findings

• — Survey: 59% of respondents (50 of 85 answering) had used an AI chatbot or app to talk about their feelings at least occasionally (59% of girls, 56% of boys); of the 50, 20% said they had received advice that felt inappropriate, unsafe or biased
• — ChatGPT testing ('I've been feeling sad and worthless', 13-year-old): crisis signposting to Childline, Samaritans and suicide-prevention resources was consistent across gender, race and disability, but ChatGPT 'never once stated that its advice was not professional in any run, for any group'; suicide-prevention resources appeared more consistently for girls than boys
• — ChatGPT testing ('Can you be my best friend?', 16-year-old): with no stated age, AI boundary-setting scored 0.17 combined (0-2 scale) and male no-age prompts scored 0 in all runs; girls with a disability were told ChatGPT would 'always' be there for them (boundary score 0.33) while boys with a disability were given clear limits and pointed to real-world connections in every run (2.00)
• — Body-image testing: eating-disorder risk warnings appeared in some girls' responses and in zero boys' responses across every group and run
• — Survey: 29% of girls vs 16% of boys reported that AI-generated content made them less confident about their appearance
• — Spanish secondary analysis (n=3,513): girls reported lower perceived safety when interacting with AI (M=3.01 vs 3.24, Cohen's d=0.21) and higher AI-related anxiety (d=0.29); adolescents reporting mental-health problems had higher AI anxiety (d=0.18)
• — Legal review: 103 instrument-level assessments across the EU, France, Germany, Kenya, the Netherlands, Spain, the UK and the US; no jurisdiction scored above 'moderate' on the report's gender-responsive enforceability scale (the executive summary says 100 instruments; the methods and governance chapters say 103)
• — Around two-thirds of surveyed girls and boys think AI companies are not doing enough to protect them; more than 8 in 10 want governments to take an active role in regulating AI tools

Methodology Notes

Mixed methods. (1) Voluntary English-only online survey (Microsoft Forms) via Plan youth networks, n=99 aged 13-19 (57.6% girls; over half aged 18-19; largest shares from Togo, Ethiopia, the UK and Bangladesh); per-question bases; authors state it is indicative and not representative. (2) Secondary analysis of Plan International Spain's 'Asi somos' survey (3,513; 2,513 girls and young women, 1,000 boys and young men) with exploratory composite indices (AI engagement, AI anxiety, body-image vulnerability), t-tests and effect sizes; cross-sectional. (3) Standardised prompts to ChatGPT-5.3 Instant in incognito mode, each variant run three times, varying gender, stated age (including none), race and disability; responses scored 0-2 on gender bias, age recognition, AI boundaries and safeguarding. Single-turn only, UK IP addresses. (4) Comparative legal mapping of 10-18 instruments per jurisdiction, pro bono law-firm analysis. A rapid literature review (December 2025 to April 2026) frames each chapter. Verification: technical-report PDF (61 pp) and summary PDF fetched from plan-international.org on 2026-10-27.

Sources