General-purpose AI has entered adolescent mental-health support, but not as a simple replacement for people or services. In a community survey of 24,132 students in England, 2.7%—662 young people—reported using an AI or large-language-model platform for mental-health support in the previous year.[1]
The proportion is small enough to resist claims that chatbots have already become the dominant source of help. The pattern within that group is more consequential. Of 580 users who rated the experience, 55.9% described it as helpful. Yet 92.1% of all reported users had also accessed another form of support, and 63.2% of those answering the question said they wanted additional help.[1]
Helpful and sufficient are different outcomes
Mental-health support in the surveyed student sample
| Survey measure | Reported result | What it can support |
|---|---|---|
| Past-year AI use for mental-health support | 662 of 24,132 students, or 2.7% | Use exists but was uncommon in the sampled population |
| Rated the support helpful | 324 of 580 respondents, or 55.9% | A majority of raters perceived some benefit |
| Also used another support source | 610 of 662 users, or 92.1% | AI usually appeared within a wider help-seeking pathway |
| Wanted additional support | 410 of 649 respondents, or 63.2% | Perceived usefulness did not remove unmet need |
Perceived helpfulness and unmet need can coexist. The survey asks students about their experiences; it does not measure symptom improvement or establish whether AI use caused them to seek more support.
The study found that both AI use and perceived need for additional support were concentrated among minoritised and more psychosocially vulnerable groups. The finding identifies who is using these tools and reporting unmet need; it does not determine whether AI reduced barriers to care or displaced better support.[1]
The product category is deliberately broad
The research concerns general-purpose AI and LLM platforms rather than a single regulated therapeutic tool. That matters because products differ in age controls, privacy practices, crisis responses, memory, tone and whether they are designed to simulate companionship. A survey answer that says “AI” cannot identify which safeguards were present or what advice was given.
Self-report creates another boundary. The data describe what students remembered and how they interpreted the interaction. They do not independently verify prompts, outputs, diagnoses, risk level or subsequent actions. The cross-sectional design also prevents a causal conclusion. Young people with greater distress may be more likely to try AI and more likely to want additional support; the technology need not have caused either condition.
Sources & further reading
- General-purpose AI as a source of mental health support amongst 24,000 adolescents in England npj Mental Health Research · 27 Oct 2026 · Accessed 30 Oct 2026