Youth Mental Health in the AI Era: How GenAI Enters Help-Seeking Pathways

Key Findings

• — Among youth experiencing mental-health struggles, 12% report using generative AI to discuss mental health concerns; of that group 69% rely only on general-purpose tools not designed for mental health support, and a further 21% use both general-purpose and mental-health-specific tools
• — Of those users, 59% reported that the system encouraged them to seek professional help or crisis services, while 41% did not report being encouraged
• — For one in three youth reporting mental-health struggles who used generative AI, it functioned as a substitute for professional care rather than a bridge to it; the remaining two in three used it before, after, or alongside professional support
• — The authors state the consequence directly: youth mental-health disclosures are occurring primarily on systems that were not designed, governed, or safeguarded for youth mental-health use
• — Prevalence of mental-health struggles is unevenly distributed: 49% among ages 18-24 versus 37% among ages 13-17, 51% of girls versus 36% of boys, 69% of LGBTQ youth versus 38% of non-LGBTQ youth, and 49% of youth experiencing financial hardship versus 37% of those not
• — Of those who used generative AI for mental health or hardship (n=112), 19% found it both helpful and unhelpful and a further share found it neutral or not helpful

Methodology Notes

Cross-sectional youth pulse survey, n=1,340 aged 13-24, fielded 8 October to 12 November 2025 by NORC using the AmeriSpeak probability panel supplemented by the opt-in Prodege panel; under-18s recruited through parent panelists with parental consent and youth assent. A youth advisory board shaped instrument design and interpretation. Several headline subgroup figures rest on small cells (the helpfulness item is n=112) and the report repeatedly cautions that confidence intervals overlap for some comparisons. Cross-sectional design cannot establish whether generative-AI use precedes or follows worse mental health. Second report in a two-part series; the first presents a segmentation framework over the same sample. Version 1, published 18 February 2026 per the report's own citation block; PDF CreationDate 2026-02-16. Retrieval note: the landing page is a JavaScript-rendered site whose PDF sits behind an embed widget, so the document is reached through the widget's CDN URL rather than a link on the page, which is why earlier sweeps recorded it as non-retrievable.

Sources