Abstract / Summary
Background: Youth-facing artificial intelligence (AI) mental health tools have moved from research prototypes to consumer products at scale, and nearly three in four (72%) US adolescents report having used an AI companion. Racial and ethnic minority and sexual and gender minority (SGM) youth carry a disproportionate mental health burden, yet how often the evidence for these tools reports who was studied has not, to our knowledge, been measured. The nearest prior review found race or ethnicity reported in 34 of 88 studies and examined gender-related indicators in none.
Objective: To measure what proportion of studies of youth-facing AI mental health tools report participants' race and ethnicity and SGM status; which groups are represented; whether outcomes are disaggregated by identity; the extent of minoritized youth involvement in design, development, or evaluation; and how reporting has changed from 2020 to 2026 and between rule-based and generative systems.
Methods: Scoping review under JBI methodology, reported against PRISMA-ScR, prospectively registered (OSF 10.17605/OSF.IO/XUZAE). Five databases were searched (2020 to 2026) with no identity term in any search string, because absence of identity reporting cannot be measured in a corpus selected on its presence. Search recall was validated against a seed set (88.6% as prespecified, 62/70; 91.1% under the operative eligibility interpretation, 82/90; 100% on the highest-priority tier under both). Identity reporting was charted for every study—absence is data.
Results: 100 studies were included. 42 of 100 (42.0%) reported neither race/ethnicity nor SGM status; 27 (27.0%) reported both. Decomposed into sexual orientation and gender identity (SOGI) data elements, race/ethnicity was reported in 42 (42.0%), gender identity in 41 (41.0%), and sexual orientation in 14 (14.0%). Outcomes were disaggregated by identity in 19 (19.0%). Involvement of minoritized youth in design, development, or evaluation was documented in 10 (10.0%), and in 0 of 36 studies of generative systems, against 6 of 40 (15.0%) rule-based conversational agent studies.
Conclusions: The evidence base for youth AI mental health tools under-reports the identities of the youth it studies; gender-identity reporting was more common in studies published from 2023 to 2026 than in studies published from 2020 to 2022, whereas sexual-orientation reporting remained uncommon in both periods; and participatory involvement of minoritized youth is unreported in the generative literature (0 of 36 studies). Reporting completeness is an actionable target for journals, funders, and developers.