AI and Study Choice
How to cite
Hertweck, Friederike; Rumert, Luis; Wolf, Rosa Ricarda; Schilling, Pia. 2026. "AI and Study Choice." AEA RCT Registry. AEARCTR-0018636. https://doi.org/10.1257/rct.18636-1.1
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
We study how high-school students form beliefs and plans about post-graduation academic programs, and how targeted information shapes these choices. Using a chatbot-based interaction system powered by agent AI using large language models (LLMs), we conduct large-scale conversational interviews with students across Germany to elicit their interests, expectations, and intended educational trajectories as well as the reasoning of those.
Embedded in the chatbot is a randomized controlled trial (RCT) to identify the causal effect of two informational treatments: (i) personalized guidance on educational and occupational pathways aligned with each student's mentioned trajectory, and (ii) concrete examples of institutions offering these and alternative pathways. This design enables us to measure whether these types of information meaningfully shift students' beliefs about their educational futures.
Intervention
Upon entering an App, the students are randomly assigned to one of three arms.
The chatbot begins with a semi-structured pre-treatment interview that is the same for all students independent of their treatment group. It walks each student through their post-graduation plans like Post-graduation plans (field of study, location preferences), reasons, perceived fit and elicits their expected returns.
1) The control group can ask questions and receives answers from verified sources, but gets no proactive information.
2) Treatment 1 additionally provides a short, general overview of the fields (of study or vocational training) the student named — typical content, key skills, entry requirements, and common career paths. Students are then encouraged to ask further questions about the given content that the chatbot will answer from verified sources.
3) Treatment 2 builds on Treatment 1 by adding information on specific study programs at German higher education institutions matching the student's interests. Students are then encouraged to ask further questions about the given content that the chatbot will answer from verified sources .
All chatbot responses draw on verified counseling sources.
Experimental design
The study is intended for high school students from academic or comprehensive tracks, that graduate with their A-levels in 2027 (Approximately 17/18 years old).
Students will download an app, whose use is preceded by the interview and intervention as well as a baseline survey. Students are randomized in one of the three treatment groups upon sign up.
Endline questions will be asked directly after intervention. After the completion of the endline survey all students will still be able to interact with the chatbot. A follow up survey in September 2027 will reveal realized study fields of participating students.
Design details
Cited by
No citation data available for this registration yet.