AI Chatbots as a Source of Political Information: Evidence from Benin

How to cite

Frezza, Eugenia Maria; Widmer, Philine; Zhuravskaya, Ekaterina; Guriev, Sergei; Eke, Joël. 2026. "AI Chatbots as a Source of Political Information: Evidence from Benin." AEA RCT Registry. AEARCTR-0018190. https://doi.org/10.1257/rct.18190-1.0

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

Eugenia Maria Frezza* , Philine Widmer , Ekaterina Zhuravskaya , Sergei Guriev , Joël Eke

* Primary investigator

Registry ID
AEARCTR-0018190
DOI
https://doi.org/10.1257/rct.18190-1.0
First registered
Last updated
Status
On going
Trial dates
2026-03-22 – 2027-06-30
Countries
Benin ()
JEL codes
D83, L86, O33, P16
Registry record
http://www.socialscienceregistry.org/trials/18190

Abstract

AI chatbots hold the promise of transforming access to information in developing countries. In Benin, there are 16.4 million mobile connections (110% of the population) but only 32% internet access. Mobile data costs are high; Sub-Saharan Africans pay on average 2.4% of monthly income for 1 GB, well above international affordability thresholds. Information search on smartphones is difficult: small screens, no tabs, and limited ability to compare or synthesize sources. Research shows that mobile-only users often have a limited "mental model" of the internet, relying on just one or two apps. In this environment, AI chatbots could be transformative: a single conversational query replaces multiple web searches and synthesizes complex information into plain language. Yet there is almost no causal evidence on the political effects of AI chatbots, and none from a low-income, developing-country context.

This project studies whether access to AI chatbots affects political knowledge, attitudes, and electoral engagement in a low-information environment. We conduct a randomized controlled trial with university students in Benin around the April 12, 2026 presidential election. We deploy a general-purpose AI chatbot via WhatsApp with one treatment arm additionally receiving access to a fact-checking chatbot. WhatsApp is the dominant communication platform across Sub-Saharan Africa, making it a natural delivery channel for our study. The core question is: Does access to AI chatbots change how people engage with political information, what they know, what they believe, and how they vote?

Intervention

We conduct a three-arm randomized controlled trial with approximately 3,000 university students recruited in person at public universities in Benin (Universite d'Abomey-Calavi and Universite de Parakou). After in-person recruitment, we follow up with participants on WhatsApp and randomly assign them to: (1) a control group with no intervention; (2) a treatment group encouraged to use a general-purpose AI chatbot deployed via WhatsApp; or (3) a treatment group encouraged to use both a fact-checking chatbot (AskVera) and the same general-purpose AI chatbot. Participants receive instructions on how to access and use the chatbots, and the cost of using the AI tools is covered by the research team. Incentivized quizzes encourage the use of these tools. Participants are free to use the tools as much or as little as they wish over the treatment period.

Experimental design

Participants are randomly assigned to one of three groups: a control group, a treatment group receiving access to a general-purpose AI chatbot via WhatsApp, or a treatment group receiving access to the same chatbot plus a fact-checking service. We collect baseline and endline survey data as well as chatbot usage logs.

Design details

Randomization method
Randomization is conducted using a computer-based random number generator in Stata. A seed is set to ensure reproducibility. Randomization stratified by gender, region of origin, and field of study (STEM vs non-STEM).
Randomization unit
Individual
Clusters
NA
Observations
3000 University Students.
Arms
1000 students will be assigned to the Control group. 1000 students will be assigned to the Treatment Arm 1. 1000 students will be assigned to the Treatment Arm 2.
Primary outcomes
- AI chatbot usage (frequency, topics discussed, user experience, trust in tool); - Political knowledge (current events quiz scores, beliefs about political issues); - Voting behavior (turnout, candidate choice); - Political attitudes (policy issue salience, policy positions, candidate familiarity and favorability); - Institutional trust and democratic satisfaction; - Self-reported well-being.

Cited by

No citation data available for this registration yet.

Keywords