1. médialab Sciences Po
  2. News
  3. A nonpartisan source-grounded AI voter guide is perceived as trustworthy and affects voting intentions

A nonpartisan source-grounded AI voter guide is perceived as trustworthy and affects voting intentions

The upcoming seminar will welcome Jonne Kamphorst, Assistant Professor in Political Science and Quantitative Social Science Methods at the CDSP and CEE. He will present his research on the use of AI tools in accessing political information and their effects on voting behavior.

Event, Research Seminar

⇥

Salle K.011, 1 place Saint Thomas d'Aquin 75007 Paris

Abstract

Generative AI tools like large language models (LLMs) are increasingly used to find and understand political information, but evidence on neutrality, accuracy, and impacts on voters remains limited. 

We developed a voter information chatbot grounded in a nonpartisan political information source (Ballotpedia) to answer questions about federal and state-level races in the 2024 U.S. general election. 

In a preregistered experiment conducted the week before the election, eligible voters in California and Texas (N = 2,474) were randomly assigned to use the chatbot or to consult their usual election information sources. 

Across party affiliations, participants rated the information from the chatbot as trustworthy, accurate, and unbiased, consistent with text analyses showing chatbot responses closely tracked  source content. Additionally, participants assigned to use the chatbot  reported higher turnout intentions, warmer affect toward supporters of the opposing party, and modest shifts in candidate vote intentions, including increased intentions to vote for Democratic candidates and, in  some races, increased intentions to vote for candidates whose positions aligned with their own. 

In a national survey of U.S. adults (N =  2,842), 13.9% reported already using AI for voter information and 49.9% reported at least some willingness to use a validated, nonpartisan AI voter guide, while most also reported concerns about accuracy and bias.  

Together, results suggest that AI voter guides grounded in nonpartisan sources could be widely adopted, can provide information voters perceive as trustworthy, and influence stated voting intentions, highlighting  the importance of transparent design and independent evaluation of the accuracy and neutrality.  

Biography

Jonne Kamphorst is an Assistant Professor in Political Science and Quantitative Social Science Methods. He is affiliated with the Centre for Socio-Political Data and the Centre for European Studies and Comparative Politics. 

Previously, he was a Postdoctoral Scholar at Stanford University’s Politics and Social Change Lab and Department of Computer Science, as well as at the European University Institute (EUI), where he received his Ph.D. in Political Science in 2023. 

His research agenda is organized around two core themes. First, he studies the politics and societies of advanced democracies, focusing on the origins of contemporary political divisions and on how democracy can be strengthened by re-engaging voters and bridging political divides. He examines these topics using quantitative scientific methods grounded in an experimental logic, including field and survey experiments and methods of causal inference. 

The second strand of his research focuses on the use of large language models in social-scientific research methods. He is particularly interested in using LLMs to simulate human behavior and as tools that can improve voters' understanding of politics. His research has been published in PNAS, the American Political Science Review, and the Journal of Politics, among other outlets.

Practical informations

This seminar will be held in person and in English, on Tuesday, March 24, 2026, from 2:00 PM to 3:30 PM, in Room K.011, 1 Place Saint-Thomas d'Aquin, 75007 Paris.

Registration is mandatory via this link.