AI-Generated Political Information and Democracy
How to cite
Petrarca, Constanza Sanhueza; Biddle, Nicholas; Chernykh, Svitlana. 2025. "AI-Generated Political Information and Democracy." AEA RCT Registry. AEARCTR-0017449. https://doi.org/10.1257/rct.17449-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
This pre-analysis plan outlines the research design and analytical strategy for a vignette survey experiment examining how the use of Artificial Intelligence affects public perceptions of election integrity, trust in investigating authorities, confidence in elections and support for democracy. The experiment is embedded within the December 2025 ANUpoll conducted online with approximately 3,500 adult Australian respondents.
Intervention
This study seeks to investigate the following research questions:
1. How and under which circumstances does AI-generated political information affect citizens’ perceptions of elections and democracy?
2. What can be done to mitigate the negative impact of AI on democracy?
The experimental design presented in this pre-analysis plan manipulates three key dimensions associated with AI to assess their effects on democratic perceptions: (1) the investigating authority, (2) the prevalence of AI-generated false information, and (3) voter ability to identify AI-generated political content.
The experiment employs a 3×3×3 factorial between-subjects design. Respondents will be randomly assigned to one of 27 experimental conditions (3× 3×3). The vignette will describe a hypothetical scenario in which the investigating authority examined the use of AI-generated content during an election campaign and reached conclusions about its prevalence and voters' ability to identify such content.
After reading the vignette, respondents are asked a series of questions measuring satisfaction with democracy, trust in findings about AI-generated political content, trust in election outcomes and perceived election legitimacy, regulatory preferences regarding AI, and concerns about comparable situations arising in Australia.
Experimental design
This study employs a 3×3×3 factorial between-subjects design (27 conditions, n≈3,500) embedded in the December 2025 ANUpoll. Respondents read a hypothetical vignette about a post-election investigation into AI-generated content, with three manipulated dimensions: (1) investigating authority (Electoral Commission/independent fact-checkers/social media platform), (2) prevalence of AI-generated false information (high/moderate/low), and (3) voter ability to identify AI content (low/moderate/high).
Design details
Cited by
No citation data available for this registration yet.