Cognitive Load and Reliance on AI Recommendations

Last registered on August 20, 2026
View Trial History

Pre-Trial

Trial Information

General Information

Title
Cognitive Load and Reliance on AI Recommendations
RCT ID
AEARCTR-0019352
Initial registration date
August 11, 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
August 20, 2026, 8:36 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
Yunshu Cao
Affiliation
Contact Primary Investigator

Other Primary Investigator(s)

Additional Trial Information

Status
In development
Start date
2026-08-11
End date
2027-02-28
Keywords
Behavior, Finance & Microfinance, Lab
Additional Keywords
Artificial Intelligence, AI Reliance, Cognitive Load, Decision Making, Human-AI Interaction
JEL code(s)
C91, D83, D91
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
This study examines how task difficulty and the correctness of AI recommendations affect decision-making performance. We plan to recruit approximately 600 participants aged 18–65, with approximately 300 participants in China and 300 in the United States, including university students and members of the general public recruited through online and in-person methods. In a 2 × 2 within-subject design, each participant completes eight reasoning and decision-making tasks accompanied by AI recommendations. Task difficulty and AI recommendation correctness are assigned at the task level according to pre-generated pseudorandom participant schedules, while the eighth task family is randomly selected at the participant level from three additional task families. The primary outcome is task-level decision accuracy, and decision confidence is the secondary outcome. The study tests whether greater cognitive demands change how individuals respond to correct and incorrect AI-provided information.
External Link(s)

Registration Citation

Citation
Cao, Yunshu. 2026. "Cognitive Load and Reliance on AI Recommendations." AEA RCT Registry. August 20. https://doi.org/10.1257/rct.19352-1.0
Sponsors & Partners
Experimental Details

Interventions

Intervention(s)
Participants will complete a series of reasoning and decision-making tasks accompanied by AI recommendations. The study experimentally varies task difficulty and whether the AI recommendation is correct or incorrect. This allows us to examine how cognitive demands and the quality of AI-provided information affect decision accuracy and confidence.
Intervention Start Date
2026-08-11
Intervention End Date
2026-11-11

Primary Outcomes

Primary Outcomes (end points)
Task-level decision accuracy
Primary Outcomes (explanation)
Task-level decision accuracy is defined as a binary indicator equal to 1 if the participant selects the objectively correct answer according to the pre-specified answer key, and 0 otherwise. Accuracy is measured separately for each core task completed by a participant.

Secondary Outcomes

Secondary Outcomes (end points)
Decision confidence
Secondary Outcomes (explanation)
Decision confidence is the participant’s self-reported confidence in the final answer after each core task, measured on a five-point scale from “not at all confident” to “extremely confident.” Confidence is conceptually distinct from perceived AI quality and will be analyzed as a secondary outcome.

Experimental Design

Experimental Design
The study uses a 2 × 2 within-subject experimental design. The two experimentally varied factors are task difficulty (easy versus difficult) and AI recommendation correctness (correct versus incorrect). Participants complete multiple reasoning and decision-making tasks, allowing the study to compare decision accuracy across different combinations of cognitive demand and AI information quality.
Experimental Design Details
Not available
Randomization Method
Randomization is implemented by computer using pre-generated pseudorandom task schedules. Seven task families are administered to all participants, and one additional task family is randomly selected from three remaining task families. Across the administered tasks, task difficulty and AI recommendation correctness vary according to the assigned schedule. For online participants, Qualtrics assigns and displays the corresponding task versions.
Randomization Unit
The primary experimental treatments—task difficulty and AI recommendation correctness—are randomized at the task level within participants. The selection of the eighth task family from the three additional task families occurs at the participant level.
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
600 participants
Sample size: planned number of observations
Approximately 600 participants, contributing up to 8 core task-level observations each (approximately 4,800 task-level observations in total).
Sample size (or number of clusters) by treatment arms
Approximately 600 participants. This is a within-subject design, so participants are not assigned to mutually exclusive treatment arms. Task difficulty (easy vs. difficult) and AI recommendation correctness (correct vs. incorrect) vary across tasks within participants.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Supporting Documents and Materials
IRB

Institutional Review Boards (IRBs)

IRB Name
University of Illinois Urbana-Champaign Institutional Review Board
IRB Approval Date
2026-06-30
IRB Approval Number
IRB26-0551
Analysis Plan

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

Request Information