Real-world use of large language models for mental health in 2024

Key Findings

• — 24% of the 1,871 surveyed US adults reported using large language models for their mental health
• — Users skew young, male and Black, and have poorer mental health than non-users
• — Reported reasons for use combine difficulty accessing traditional treatment with the models being free, convenient and available
• — Reported uses are emotional support, learning therapy skills, and supplementing existing therapy
• — Applying Pew population LLM-use estimates, the authors conservatively estimate 14-18 million US adults may have been using LLMs for mental health as of 2024

Methodology Notes

Cross-sectional survey, n=1,871 US adults, quota/stratified sampling on age, sex and race/ethnicity to approximate national demographics; self-report. The most important caveat for citation is the field date: data were collected August-October 2024 and published 2026-08-01, so the 24% figure describes the 2024 population and predates much of the companion-app growth since. The 14-18 million extrapolation is the authors' own, derived by applying Pew LLM-use estimates to their quota sample, and rests on the stated presumption that the quota sample fairly approximates the population. Published in npj Digital Medicine under CC BY-NC-ND 4.0 (note: more restrictive than the CC BY typical of this journal). Verified by fetching nature.com directly with a browser user agent (abstract and the 14-18 million passage read in the article body) and cross-checked against Crossref (title, authors, 2026-08-01 online date, licence) and PubMed PMID 42601385.

Sources