Teacher-Designed AI for Personalized Learning: Evidence from Roma Students in Spain

How to cite

Navarro-Sola, Laia; Cobreros, Lucía; Roldán-Monés, Antonio. 2025. "Teacher-Designed AI for Personalized Learning: Evidence from Roma Students in Spain." AEA RCT Registry. AEARCTR-0017234. https://doi.org/10.1257/rct.17234-3.0

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

Laia Navarro-Sola* , Lucía Cobreros , Antonio Roldán-Monés

* Primary investigator

Registry ID
AEARCTR-0017234
DOI
https://doi.org/10.1257/rct.17234-3.0
First registered
Last updated
Status
In development
Trial dates
2025-11-16 – 2027-03-30
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17234

Abstract

This study evaluates whether empowering teachers to design and use artificial intelligence (AI) tutoring tools can improve student learning and engagement in after-school educational programs in Spain. Despite sustained investment in remedial education, teachers continue to face challenges in tailoring instruction to students’ diverse learning levels and needs. Through a no-code platform, teachers in treatment centers receive training and coaching to create and adapt AI tutoring chatbots aligned with curricular goals and student learning profiles. The chatbots provide adaptive practice and feedback during after-school sessions, complementing, rather than replacing, human instruction. The study will use a cluster-randomized controlled trial across multiple sites and regions in Spain, with randomization at the teacher-group level to prevent spillovers. It will measure impacts on student achievement, as well as engagement and effort, and self-efficacy.

Intervention

Educators in after-school programs for Roma students in Spain receive structured training, ongoing coaching, and access to a no-code platform that allows them to design and adapt AI tutoring chatbots for their students. Through the platform, teachers specify prompts, materials, and pedagogical rules so that chatbots can deliver guided practice and feedback aligned with curricular goals and students’ needs, preferences, and identities. Teachers in the control group continue standard after-school instruction without training or platform access.

Experimental design

Our baseline sample consists of approximately 1,700 students enrolled in after-school programs across 60 centers in 13 regions of Spain. The study randomizes 120 teacher-block clusters, defined as one or more educators who co-teach or share students within the same group.

Clusters are randomly assigned with equal probability to treatment or control. Treatment teacher-blocks receive structured training, ongoing coaching, and access to a no-code AI platform that allows them to design and adapt tutoring chatbots for their students. Control teacher-blocks continue standard after-school programming without access to the platform or training. Randomization is stratified by after-school program type, site or region, and baseline cluster achievement.

Our main empirical specifications will estimate intention-to-treat effects, reflecting the causal impact of assignment to each treatment arm on our outcomes of interest.

Design details

Randomization method
Randomization done in office by a computer
Randomization unit
Clustered randomization occurs at the teacher-block level, defined as one or more educators who co-teach or share students within the same after-school group, to prevent spillovers between teachers working with the same students.
Clusters
Approximately 120 teacher-block clusters
Observations
Approximately 1,700 students across 60 centers in 13 regions in Spain
Arms
60 teacher-block clusters treatment, 60 teacher-block clusters control
Primary outcomes
Standardized test scores in math, reading, and English
Secondary outcomes
Student-level outcomes: Engagement, self-efficacy, trust in AI, and sense of belonging Teacher-level outcomes: Self-efficacy, stress, time use, and adoption of personalized pedagogical strategies

Cited by

No citation data available for this registration yet.

Keywords