General-purpose AI as a source of mental health support amongst 24,000 adolescents in England

Key Findings

• — Of 24,132 students who reached the relevant survey section, 662 (2.7%) reported past-year use of AI/LLM platforms for mental health support
• — Of users who rated it, 55.9% (324/580) found the support helpful, 31.4% neither helpful nor unhelpful and 12.8% (74/580) unhelpful
• — 63.2% (410/649) of users reported wanting additional mental health support; 92.1% (610/662) had also accessed other forms of support and 46.1% (305/662) had accessed formal services
• — 16.5% of users used only online sources for support, around half of whom (n = 52) used only AI/LLMs
• — Use was more likely among trans or gender-diverse students (aOR 2.24, 95% CI 1.61-3.09), those who had ever self-harmed (aOR 1.39, 95% CI 1.09-1.78), those often lonely (aOR 1.58, 95% CI 1.20-2.10) and those with higher depression/anxiety scores (aOR 1.46 per SD on the RCADS-25)
• — Among users, having ever self-harmed was associated with perceived unmet need for additional support (aOR 2.12, 95% CI 1.28-3.57)

Methodology Notes

Repeated cross-sectional school survey (OxWell 2025 wave), responses collected February-May 2025 in schools mainly in Oxfordshire and Liverpool; non-probabilistic school recruitment via local authorities, a multi-academy trust and an FE college group; parental opt-out with student assent under 16. 36,169 students logged in, 32,102 consented, 24,132 reached the AI question; mean age 14.0. AI use was a single item listing 'AI/LLM platforms (e.g. ChatGPT, Gemini, Copilot)' among support sources; helpfulness a three-point item. Complete-case logistic regression (n = 22,659 for use). No regression for substitution because only 52 students used AI alone. The discussion text gives 64.1% wanting additional support while the results and abstract give 63.2% (410/649); the latter is used here. Published as an accelerated 'Article in Press' (received 6 March 2026, accepted 5 October 2026, online 27 October 2026); the HTML page shows only the abstract, and the full text was read from the article-in-press PDF on nature.com. Funded by the NIHR Applied Research Collaboration Oxford and Thames Valley, with support from Liverpool City Council and Oxfordshire County Council.

Sources