Effects of Fact-Checker Label and National Stereotypes on Perceived News Credibility

How to cite

Casanova, Jonathan; Tavares, Jose Albuquerque. 2025. "Effects of Fact-Checker Label and National Stereotypes on Perceived News Credibility." AEA RCT Registry. AEARCTR-0016950. https://doi.org/10.1257/rct.16950-1.0

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

Jonathan Casanova* , Jose Albuquerque Tavares

* Primary investigator

Registry ID
AEARCTR-0016950
DOI
https://doi.org/10.1257/rct.16950-1.0
First registered
Last updated
Status
In development
Trial dates
2025-10-16 – 2025-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16950

Abstract

This study investigates whether people trust fact-checks differently depending on whether they are labeled as coming from AI or from humans, and how this interacts with preconceptions about corruption. In a survey experiment, participants are randomly assigned to read a real news story about corruption in either Sweden or Rwanda, with each story designed to either confirm or challenge common expectations. Participants first report how believable they find the story. They are then informed that the story was fact-checked and confirmed as true. Further, they are informed the fact-checking was conducted by an AI system or, alternatively, by a team of human fact-checkers. Participants then rate their trust in the fact-check. The study follows a between-subjects design to test how source and content influence perception. Primary outcomes include belief in the story and trust in the fact-check. Secondary outcomes include familiarity with the story, perceived alignment with expectations, and familiarity with AI. This research aims to better understand how source labels and national stereotypes affect trust in fact-checking.

Intervention

Participants are exposed to a real news story about corruption in either Sweden or Rwanda. The story is designed to either confirm or challenge common expectations about corruption in that country. After reading, participants are told the story was fact-checked and confirmed as true, with the source randomly attributed to either an AI system or human fact-checkers.

Experimental design

This study uses a between-subjects survey experiment with random assignment to one of eight conditions varying in story country (Sweden or Rwanda), story type (confirming or disconfirming expectations), and fact-checker label (AI or human). Participants are recruited from a university student population and complete the survey online.

Design details

Randomization method
Randomization is automated through Qualtrics' built-in randomizer feature. Assignment to experimental conditions occurs upon survey entry, ensuring equal probability across the eight conditions.
Randomization unit
The unit of randomization is the individual participant.
Clusters
None — individual-level randomization.
Observations
Approximately 320 participants, all university students or similar respondents, recruited through classroom sessions or online channels.
Arms
The study uses 8 experimental groups (2x2x2 design), with approximately 40 participants per condition.
Primary outcomes
• Belief in the news story (measured on a Likert scale). • Trust in the fact-check confirmation (measured on a Likert scale).
Secondary outcomes
• Familiarity with the topic under scrutiny • Familiarity with AI systems

Cited by

No citation data available for this registration yet.

Keywords