Aligned to Protect, Misaligned to Advise: Financial Advice from LLMs

Last registered on September 21, 2026
View Trial History

Pre-Trial

Trial Information

General Information

Title
Aligned to Protect, Misaligned to Advise: Financial Advice from LLMs
RCT ID
AEARCTR-0019679
Initial registration date
September 09, 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, 8:08 AM EDT

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

Locations

Country
United States of America
Region

Primary Investigator

Name
Saurabh Bhargava
Affiliation
Contact Primary Investigator

Other Primary Investigator(s)

Additional Trial Information

Status
Completed
Start date
2026-09-09
End date
2026-09-10
Keywords
Behavior
Additional Keywords
JEL code(s)
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
This experiment is part of a broader study examining how large language models provide financial advice and whether their efforts to interpret user preferences and protect users distort their recommendations. The broader project combines administrative insurance choices, Reddit requests for advice, and controlled experiments involving insurance and retirement decisions. This study will compare retirement allocation recommendations from 350 adults recruited through Prolific with previously collected model recommendations. Participants will advise two hypothetical investors whose ages differ by ten years and whose other financial circumstances are identical, recommending how to divide retirement savings between diversified stock and bond funds. We will examine how human and model recommendations vary with investor circumstances and compare them with an institutional retirement glide path.
External Link(s)

Registration Citation

Citation
Bhargava, Saurabh. 2026. "Aligned to Protect, Misaligned to Advise: Financial Advice from LLMs." AEA RCT Registry. September 21. https://doi.org/10.1257/rct.19679-1.0
Sponsors & Partners
Experimental Details

Interventions

Intervention(s)
Intervention (Hidden)
Intervention Start Date
2026-09-09
Intervention End Date
2026-09-10

Primary Outcomes

Primary Outcomes (end points)
The primary outcome is the recommended percentage of the hypothetical investor’s current retirement savings allocated to a diversified stock fund, measured as a whole number from 0 to 100.

We will examine the average recommended equity share and its sensitivity to investor age, expressed as the change in recommended equity share per ten-year increase in age. We will compare these outcomes with recommendations from three language models, an institutional retirement glide path, and optimal allocations implied by a theoretical portfolio-choice model under prespecified assumptions.

Primary analyses will use each participant’s first recommendation.
Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
We will conduct an approximately five-minute online survey of 350 adults recruited through Prolific and administered through Qualtrics. Participants will be randomly assigned to a pair of hypothetical investor profiles, presented one at a time. The investors’ ages will differ by ten years, while their other financial circumstances will be identical within each pair.

Across pairs, scenarios will vary investor age and household income. Each profile will describe retirement savings, planned contributions, retirement age, and liquidity needs. Participants will recommend the percentage of each investor’s current retirement savings to invest in a diversified stock fund, with the remainder invested in a diversified bond fund. Recommendations will be entered as whole numbers from 0 to 100, and the investor will be permitted to revise the allocation in future years.

The design will allow comparisons of retirement recommendations across investor circumstances, within participants, and between human participants and language models.
Experimental Design Details
Randomization Method
Randomization done by qualtrics.
Randomization Unit
Individual.
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
350
Sample size: planned number of observations
350
Sample size (or number of clusters) by treatment arms
350
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Supporting Documents and Materials
IRB

Institutional Review Boards (IRBs)

IRB Name
Oberlin College
IRB Approval Date
2026-08-03
IRB Approval Number
#AY26-27-TH-01
Analysis Plan

Post-Trial

Post Trial Information

Study Withdrawal

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

Request Information

Intervention

Is the intervention completed?
No
Data Collection Complete
Data Publication

Data Publication

Is public data available?
No

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials