Artificial Teaching at the Right Level: Evidence from AI-enhanced Learning Platform in Higher Education

How to cite

Grigaitis, Justinas; Braun, Tabea. 2026. "Artificial Teaching at the Right Level: Evidence from AI-enhanced Learning Platform in Higher Education." AEA RCT Registry. AEARCTR-0018206. https://doi.org/10.1257/rct.18206-2.0

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

Justinas Grigaitis* , Tabea Braun

* Primary investigator

Registry ID
AEARCTR-0018206
DOI
https://doi.org/10.1257/rct.18206-2.0
First registered
Last updated
Status
In development
Trial dates
2026-03-25 – 2026-06-04
Countries
Private
JEL codes
I24; I23; C93; D83; O33
Registry record
http://www.socialscienceregistry.org/trials/18206

Abstract

This study evaluates how voluntary AI-assisted practice tools affect student outcomes in higher education. We run a field experiment in a mandatory economics course at a large research university. All enrolled students (target sample of approx. 550) are offered access to a course-specific online learning platform with AI-assisted self-study tools. Impact evaluation relies on variation in AI feature availability across parts of the course and individual-level randomization into alternative default AI tool configurations. Primary outcomes are exam performance as well as engagement with AI-tools and traditional course materials.

Intervention

Provision of course-specific educational AI tools for voluntary AI-assisted study.

Experimental design

This study evaluates AI-assisted self-study tools in a mandatory undergraduate economics course at a Swiss research university. All enrolled students are offered voluntary access to a course-specific online learning platform. The platform contains multiple AI-assisted practice tools (e.g., chatbot and quizzes with AI feedback), which rely on official course materials (e.g., lecture slides and transcripts). The study uses two empirical components: (i) variation in AI feature availability throughout the semester, and (ii) individual-level randomization into alternative default AI tool configurations. Randomization is at the student level (no clustering). The target sample is approximately 584 enrolled students.

Design details

Randomization method
Treatment group is based on a random encrypted user ID (hash) assigned upon login.
Randomization unit
Individual
Clusters
582 individuals
Observations
582 individuals
Arms
294 individuals in the more difficult default AI content version; 288 individuals in the less difficult default AI content version.
Primary outcomes
Exam performance; Usage of digital learning platform; Self-reported and observed online usage of official study materials (lecture attendance, self-learning video watching, practice exercises, studying with peers); Endline course-specific knowledge quiz a week before the exam.
Secondary outcomes
Text data from student-AI interactions.

Cited by

No citation data available for this registration yet.

Keywords