Equalizer or Amplifier? Feasibility pilot for AI-assisted university examinations

Last registered on September 21, 2026
View Trial History

Pre-Trial

Trial Information

General Information

Title
Equalizer or Amplifier? Feasibility pilot for AI-assisted university examinations
RCT ID
AEARCTR-0019746
Initial registration date
September 16, 2026

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
September 21, 2026, 10:26 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Name
ANDREA CARLO MICHELE ICHINO
Affiliation
University of Bologna and European University Institute
Contact Primary Investigator

Other Primary Investigator(s)

PI Name
Chiara Fumagalli
PI Affiliation
Bocconi University
Contact Investigator
PI Name
Myriam Mariani
PI Affiliation
Bocconi University
Contact Investigator

Additional Trial Information

Status
In development
Start date
2026-09-17
End date
2027-01-31
Keywords
Education, Labor
Additional Keywords
generative AI, higher education, examinations, cognitive ability, learning, inequality
JEL code(s)
C93, I23, O33
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
This individually randomized feasibility pilot tests the implementation of assigned access to a university-provided generative artificial intelligence (AI) tool during formal assessments in one undergraduate course. After the consent deadline, students who have provided informed consent are included in the fixed randomization roster and assigned in approximately equal numbers to two sequences. Sequence G1 receives AI access for the Part A assessment and no AI access for the Part B assessment; G2 receives the reverse sequence. Both groups then complete a common cumulative assessment without AI. Students who do not consent are not randomized, complete all assessments without AI under the examination pathways available in the course, and are excluded from the research dataset. Primary outcomes concern consent, assessment completion, technical implementation, and research-data completeness. Academic-performance contrasts are exploratory. The pilot will inform the protocol and preregistration of a separate, larger main experiment.
External Link(s)

Registration Citation

Citation
Fumagalli, Chiara , ANDREA CARLO MICHELE ICHINO and Myriam Mariani. 2026. "Equalizer or Amplifier? Feasibility pilot for AI-assisted university examinations." AEA RCT Registry. September 21. https://doi.org/10.1257/rct.19746-1.0
Sponsors & Partners
Experimental Details

Interventions

Intervention(s)
The intervention is assigned access to a university-provided generative artificial intelligence (AI) tool during one part-specific course assessment. G1 has AI access for the Part A assessment and no AI access for the Part B assessment. G2 has no AI access for Part A and AI access for Part B. Immediately after the Part B assessment, both groups complete a cumulative assessment on Parts A and B without AI and without receiving feedback or solutions between the two assessments. Students assigned AI access may choose not to use the tool. The main comparisons follow assignment rather than actual use.
Intervention Start Date
2026-10-20
Intervention End Date
2027-01-20

Primary Outcomes

Primary Outcomes (end points)
1. Research-consent rate among eligible students.
2. Completion rates for the Part A, Part B, and cumulative assessments among randomized participants, by assigned sequence.
3. Technical implementation rate in AI-permitted assessments: the share of randomized participants who sit the assessment and for whom assigned access and required capture of the dedicated AI interaction are completed without a recorded technical failure.
4. Research-data completeness: the share of randomized participants who sit an assessment and have the assignment, assessment score, required baseline data, and, when applicable, AI-interaction record available for analysis.
Primary Outcomes (explanation)
The pilot outcomes will be reported as counts and proportions, overall and by assigned sequence where relevant. Technical failures and departures from the planned procedure will also be described by type. The pilot has no confirmatory treatment-effect hypothesis and no single numerical success threshold.

Secondary Outcomes

Secondary Outcomes (end points)
Exploratory intention-to-treat contrasts between G1 and G2 in Part A performance at the midterm and in the Part A component of the later cumulative assessment, including mean performance, score dispersion, and their relationship with the admission-test score. AI-interaction measures will be descriptive. Brief survey measures, if collected, will be analyzed descriptively only after separate approval of the survey materials and corresponding consent additions by the Bocconi Research Ethics Committee.
Secondary Outcomes (explanation)
Pilot performance results will be labeled exploratory, kept separate from the main experiment, and used only to inform the later protocol and preregistration. Analyses will not condition on actual AI use.

Experimental Design

Experimental Design
The pilot is conducted in one undergraduate course at Bocconi University. All students on the eligible course roster receive the participant information and decide whether to consent before randomization. After the consent deadline, the fixed roster of consenting students is randomized individually and in approximately equal numbers to G1 or G2. Only these students enter the experimental assessment procedure. Students who do not consent are not randomized; they may choose among the examination pathways available in the course, complete all assessments without AI, and are excluded from the research dataset. Randomized students following the partial-assessment pathway receive their assigned sequence. Graders are blinded to sequence assignment when grading the three assessment components. Research analyses include randomized participants with the outcome required for the relevant analysis.
Experimental Design Details
Not available
Randomization Method
After the consent deadline, a reproducible computer randomization assigns the fixed roster of consenting students to G1 or G2 in numbers as equal as possible. The code, seed, input roster, date, and assignment output are archived before assignments are communicated.
Randomization Unit
Individual student
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
Not applicable; assignment is at the individual-student level.
Sample size: planned number of observations
All students enrolled in the eligible course will be invited to consent; the course roster is expected to contain approximately 130 students. The randomized sample will consist of all students who provide informed consent by the prespecified deadline.
Sample size (or number of clusters) by treatment arms
All consenting students will be randomized. G1 and G2 will contain numbers as equal as possible: if the number of consenting students is even, each arm will contain half; if it is odd, one arm will contain one more student than the other. G1 has AI for Part A and no AI for Part B; G2 has the reverse sequence.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
The pilot uses the fixed enrolment of one course and is designed to assess feasibility rather than to detect a prespecified academic-performance effect. No confirmatory minimum detectable effect is claimed for the pilot.
Supporting Documents and Materials
IRB

Institutional Review Boards (IRBs)

IRB Name
Bocconi Research Ethics Committee
IRB Approval Date
2026-09-16
IRB Approval Number
RA001263
Analysis Plan