AI Support for Human Tutors to Address Latin America’s Teacher Shortage

How to cite

Llanes, Salome Aguilar; Salazar, Andres; Prakash, Alok. 2025. "AI Support for Human Tutors to Address Latin America’s Teacher Shortage." AEA RCT Registry. AEARCTR-0016870. https://doi.org/10.1257/rct.16870-1.0

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

Salome Aguilar Llanes* , Andres Salazar , Alok Prakash

* Primary investigator

Registry ID
AEARCTR-0016870
DOI
https://doi.org/10.1257/rct.16870-1.0
First registered
Last updated
Status
In development
Trial dates
2025-10-06 – 2026-01-12
Countries
Private
JEL codes
J24, I25
Registry record
http://www.socialscienceregistry.org/trials/16870

Abstract

We conduct a randomized controlled trial (RCT) of an online math tutoring program in Peru to evaluate whether AI tools can augment human tutors. Our study investigates whether AI can (i) lower barriers to recruiting volunteer tutors, (ii) improve tutoring quality as measured by student learning outcomes and perceptions, (iii) increase tutor retention, and (iv) affect both tutor and student motivation. In parallel, we systematically map tutor tasks to identify which activities are most amenable to AI support. Unlike approaches that seek to replace human teachers with AI, our intervention emphasizes AI as a complement to human tutoring, enhancing communication and instructional effectiveness. Beyond estimating causal effects, our study generates practical insights on best practices for AI-assisted math tutoring and provides a repository of tutor- and AI-generated instructional content, offering a scalable model for addressing teacher shortages in resource-constrained settings.

Experimental design

Tutors are volunteer university students recruited from a university in Peru. Information about the tutoring opportunity will be disseminated through the university’s communication channels. Tutors will indicate interest in participating in the RCT by signing up to receive additional information.

Stage 1: Initial Randomization (AI Information Exposure)

Tutors will be randomly assigned to one of two groups:
AI-Information Group: Tutors are informed that they will receive training to prepare their classes, as well as training in AI tools to assist in tutoring activities. All tutors in this group will receive the AI tools training.

No-AI Information Group: Tutors are informed that they will receive training to prepare their classes, with no mention of AI tools.

We will first compare conversion rates, which is the proportion of tutors who sign up and start teaching—between these two groups.
Stage 2: Student and Tutor Assignment
Students will be randomly assigned to groups, and groups will in turn be randomly assigned to tutors. All tutors will begin teaching math classes.

Stage 3: Nested Randomization of AI Training
Tutors in the No-AI Information Group will undergo a second randomization into two subgroups:
No-AI Training: Tutors receive standard tutor training with no explicit mention of AI.

AI Training: Tutors receive a version of the standard training that incorporates instruction in AI tools and AI literacy.

We will compare outcomes—including student learning, student perceptions, tutor retention, and tutor/student motivation—between these two subgroups and against the AI-Information Group.

Covariates
All analyses will control for tutor and student demographic variables, including university major and campus (the university has several campuses).

Was the treatment clustered?
The main outcomes won’t be clustered. The randomization will be done at the tutor level. The math learning outcomes will be clustered at the tutor level.

Subgroup analysis will be performed by gender and major, by baseline differences in teaching styles and quality.

Design details

Randomization method
Randomization done in office by computer
Randomization unit
Individual for tutors and cluster-level for the students
Clusters
Students will be randomized into groups of about 5, and will be randomly assigned to tutors. Tutors will be randomly assigned to treatment and control groups. This means we plan to have 1000 clusters.
Observations
Tutors: 1000 Students: 5000
Arms
Three treatment arms. AI-Information Group: Tutors are informed that they will receive training to prepare their classes (1/3tutors) AI Training: Tutors receive a version of the standard training that incorporates instruction in AI tools and AI literacy (1/3 tutors) No-AI Training: Tutors receive standard tutor training with no explicit mention of AI(1/3 tutors)
Primary outcomes
AI literacy questionnaire Math evaluation for students to assess quality of teaching Tutor evaluations by students Tutor retention Tutor conversion from expressing interest to actually participating Lesson plans to assess quality of teaching Class recordings to assess quality of teaching Mapping tutor tasks to prompts

Cited by

No citation data available for this registration yet.

Keywords