Can AI Guidance Improve University Programme Decisions?

How to cite

Kohler, Benjamin. 2026. "Can AI Guidance Improve University Programme Decisions?." AEA RCT Registry. AEARCTR-0018832. https://doi.org/10.1257/rct.18832-1.0

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

Benjamin Kohler*

* Primary investigator

Registry ID
AEARCTR-0018832
DOI
https://doi.org/10.1257/rct.18832-1.0
First registered
Last updated
Status
In development
Trial dates
2026-06-10 – 2026-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18832

Abstract

This pre-analysis plan describes the design and planned empirical strategy for a randomised controlled trial (RCT) studying the causal effects of AI-powered guidance on university programme decisions among Danish higher-education applicants. We randomly assign applicants to an AI chatbot that provides personalised information about Danish university programmes or to the standard national information portal ug.dk. We measure effects on stated programme preferences, decision confidence, perceived informedness, decision stress, and --- via administrative registry linkage --- actual application choices and dropout rates.

Intervention

Participants are randomly assigned to one of two tools to support their university programme decision. The treatment group receives access to a custom AI chatbot, that provides personalised information about Danish higher education programmes. The chatbot is augmented with information retrieval tools for programmes and their institution-specific offerings; all underlying information is publicly available on ug.dk, and all conversation logs are stored and linked to the participant. The control group receives access to the standard national information portal ug.dk, i.e., the current state of the art. Both groups interact with their assigned tool within the session and are encouraged to keep using it until the national admission deadline.

Experimental design

A two-arm, individual-level randomised controlled trial. Within a single online session, participants complete a pre-survey, are randomly assigned to treatment (AI chatbot) or control (ug.dk), interact with their assigned tool, and complete a first post-survey (Post-survey I). Participants are encouraged to keep using their tool after the session. One day after the national higher-education admission deadline (6 July 2026) — they are re-contacted for Post-survey II. Administrative registry data are linked via CPR number after Post-survey II, and a third follow-up survey is administered after the first university semester (collecting well-being, programme satisfaction, and any subsequent switching). Participation is incentivised through a raffle of 200 gift cards of 1,000 DKK, with a doubled raffle entry for participants who spend at least three minutes with their tool.

Design details

Randomization method
Randomisation is pre-assigned at the invitation stage. This is done by the research team in office (by a computer) before invitations are sent, ensuring exact 50/50 balance and preventing within-session switching.
Randomization unit
Individual (participant level).
Clusters
No clustering — randomisation is at the individual level. (Equivalently, the number of clusters equals the number of individual participants, ~2,000.)
Observations
Approximately 2,000 participants. The sampling frame covers ~40% of each birth cohort aged 19–21 from Statistics Denmark administrative records; at an anticipated 5% response rate, we expect roughly 2,000 participants.
Arms
50/50 allocation: approximately 1,000 in the treatment arm (AI chatbot) and approximately 1,000 in the control arm (ug.dk).
Primary outcomes
Primary outcome groups: 1. Application decision — (a) whether the top-ranked programme changes between the pre-survey and each post-survey (binary revision indicator); (b) rank-order correlation between pre-survey and post-survey rankings across all stated choices; (c) whether previously unlisted programmes are newly chosen post-treatment. 2. Application propensity — binary indicator for whether the participant submitted a formal application to the Coordinated Enrolment system (KOT) by the July deadline. 3. Knowledge — (a) perceived informedness (self-reported knowledge about programme options, Likert); (b) admission realism (absolute gap between self-assessed admission probability for the first-choice programme and both the actual GPA cut-off and the model-estimated admission likelihood following Agarwal & Somaini 2018). 4. Well-being — (a) decision confidence (certainty about the top-ranked programme); (b) application-induced stress. More details in the PAP.
Secondary outcomes
See PAP for details.

Cited by

No citation data available for this registration yet.

Keywords