AI Mental Health Take-up and Social Demographic Background

How to cite

Du, Jiachuan; Meng, Juanjuan; Gangadharan, Lata; Gao, Yu; Bao, Leo. 2026. "AI Mental Health Take-up and Social Demographic Background." AEA RCT Registry. AEARCTR-0018387. https://doi.org/10.1257/rct.18387-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Jiachuan Du* , Juanjuan Meng , Lata Gangadharan , Yu Gao , Leo Bao

* Primary investigator

Registry ID
AEARCTR-0018387
DOI
https://doi.org/10.1257/rct.18387-1.0
First registered
Last updated
Status
In development
Trial dates
2026-05-02 – 2026-08-02
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18387

Abstract

This study examines whether AI mental health services can reduce socioeconomic inequality in access to mental health support, relative to traditional human mental health services.

Mental health services are often underutilized, particularly among individuals from lower socioeconomic backgrounds. Financial constraints, stigma, accessibility barriers, and concerns about judgment may disproportionately affect these groups. AI-based counseling tools may lower some of these barriers by offering lower cost, greater immediacy, and perceived anonymity.

In an online experimental survey, participants are introduced to both AI-based and human-provided mental health services. We elicit their willingness to pay (WTP) using the Becker-DeGroot-Marschak (BDM) incentive-compatible mechanism and randomly implement real purchase opportunities for a subset of participants. The study collects detailed information on socioeconomic and demographic characteristics, prior AI usage, prior mental health help-seeking, and channel questions that explains take-up behavior.

In addition to immediate take-up decisions, we conduct a one-month follow-up survey to measure subsequent real-world (field) utilization of AI and human mental health services.

The primary objective is to assess whether the gap in demand and take-up between higher and lower socioeconomic groups is smaller for AI services than for human services. By combining BDM-based valuation with behavioral take-up decisions, the study aims to provide evidence on whether AI can reduce inequality in mental health support take-up.

Experimental design

Participants complete an online survey that includes mental health screening, socioeconomic and demographic characteristics, prior AI usage, prior mental health help-seeking, and channel questions that explains take-up behavior.

All participants are introduced to both an AI-based and a human-delivered counseling service, with order randomized. After each introduction, participants report their maximum willingness to pay for a 30-minute session using the Becker-DeGroot-Marschak (BDM) mechanism.

A subset of participants is randomly selected to receive a monetary endowment and is further randomly assigned to face a real purchase decision for either the AI or human service. A price is randomly drawn according to the BDM procedure and compared to the participant’s stated WTP to determine whether the service is provided.

Approximately one month after the experiment, all participants are invited to complete a follow-up survey. The follow-up survey measures subsequent real-world use of AI mental health services and human mental health services during the one month period.

Design details

Randomization method
Randomization is implemented using the built-in randomization and random number generation functions of the online survey platform.
Randomization unit
Randomization occurs at the individual level for: The order of service demonstration presentation (AI first vs. human first), Assignment to receive a monetary endowment, Assignment to AI versus human service for real implementation under the Becker-DeGroot-Marschak (BDM) mechanism, and The random price draw in the BDM mechanism.
Clusters
2,000 individuals. (This number represents the anticipated maximum feasible sample size rather than a guaranteed enrollment, the achieved sample may be lower depending on field conditions. Power analysis based on pilot data shows that the minimum sample size required to detect the relevant effects with adequate power is roughly 1,800.)
Observations
2,000 individuals.
Arms
1,000 assigned to AI BDM implementation; 1,000 assigned to human BDM implementation. At recruitment, the sample was stratified to ensure balance across three demographic characteristics: gender (50% male, 50% female), hukou status (50% urban hukou, 50% rural hukou), and individual monthly income (50% RMB 5,000 or below, 50% above RMB 5,000).
Primary outcomes
Willingness to pay (WTP) for AI mental health services. Willingness to pay (WTP) for human mental health services.

Cited by

No citation data available for this registration yet.

Keywords