Visual Prebunking Advertisements Perform Better Than Their Audio-Only Counterpart for Improving Information Literacy

Abstract

A Review of: 

Daly, D., & Jarrette, K. (2025). Design of audio ads to prebunk misinformation and promote civil discourse. Information Research: An International Electronic Journal, 30(iConf), 249–259. https://doi.org/10.47989/ir30iConf47359 

Objective – To determine if the use of prebunking advertisements influences information literacy, the ability to identify false news headlines, or attitudes toward civil discourse. 

Design – A pilot, experimental study. 

Setting – A large university in the southwestern United States. 

Subjects – 143 undergraduate students. 

Methods – A research team developed five short audio advertisements intended for prebunking sources of misinformation identified through social media. For each misinformation strategy, the team created a humorous sketch, dramatizing an interaction between two characters who knew each other. The team created familiar characters to model how one could engage friends or family who could be susceptible to believing misinformation and promote civil discourse among them. The audio ads were intended to be aired during podcasts known to spread misinformation.  

For the experimental design, the audio ads were coupled with Artificial Intelligence (AI)-generated visualization. Researchers set out to determine whether exposure to a specific prebunking ad enhances an individual’s ability to identify false news headlines, whether the visualization of the ad script using AI assistance impacts respondent literacy, and how participants describe and gauge the effectiveness of a specific prebunking audio ad.   

Participants were recruited through instructors who taught courses related to study topics. Instructors were encouraged to offer extra credit for participation. In Part 1 of the study, participants answered questions about demographics and social media use.   

Participants completed two established qualitative questionnaires: the Generic Conspiracist Beliefs scale (GCBS) and the Misinformation Susceptibility Test (MIST-20) (Maertens et al., 2024). The researchers developed a questionnaire modeled after the MIST-20, the ITMIST, using real and fake headlines. Participants were exposed to one ad: either an audio-only ad, an AI-generated visualization ad, or a control ad. Participants completed another qualitative questionnaire after viewing the ad to finish Part 1. The following day, participants received a link to complete the GCBS, MIST-20, and ITMIST and completed another qualitative questionnaire within a week of the first survey, to finish Part 2 of the study. 

Main Results – One hundred forty-three participants completed Part 1 of the study, and 99 completed Part 2. Participants ranged in age from 18–48 years; 59.6% identified as female, 38.4% identified as male; 54.5% identified as White/Caucasian, with the remaining participants identifying as racially diverse; 34.4% identified as Democrat, 32.3% Republican, 18.2% Independent; and participants represented multiple religious affiliations.  

All participants used a social media platform at least once a week: 43.4% reported usage over two hours per day, 26.3% between 90–120 minutes, 12.1% between 60–90 minutes, 14.1% between 30–60 minutes, and 4% less than 30 minutes. Nearly 90 percent (89.9) of participants used Instagram, 67.6% TikTok, 66.7% Snapchat, 34.3% Twitter/X, 21.2% Facebook, and 8.1% used other social media platforms. Regarding podcasts, 23.2% frequently tuned in, 50.5% sometimes tuned in, and 26.3% never tuned in. Of those who listened to podcasts, 71.2% always skipped podcast ads, 26% sometimes skipped, and 1.4% never skipped. The podcasts that participants reported frequently tuning into for entertainment and education were strongly related to stated political affiliation. 

The authors reported the results of the MIST-20 and ITMIST in this article. At the time of publication, the authors were still analyzing the results of the GCBS and the complete quantitative and qualitative data. When comparing the AI-generated visualization ad (Visual Experimental group) to the Visual Control group, investigators reported a significantly large average improvement in information literacy scores for the Experimental group on the MIST-20 (Visual Experimental x̄ = 0.93, Visual Control x̄ = 0.33), and a moderate average improvement on the ITMIST (Visual Experimental x̄ = 0.98, Visual Control x̄ = 0.81). When comparing the Audio Experimental group to the Audio Control group, investigators report mixed results. The Audio Experimental group did not show as great an average improvement compared to the Control group on the MIST-20 (Audio Experimental x̄ = 0.85, Audio Control x̄ = 1.41) but scored higher than the Control group on the ITMIST (Audio Experimental x̄ = 0.78, Audio Control x̄ = 0.45). More than half of the participants in each Experimental group improved in score. Those who improved showed a greater change in score than those whose score declined. 

Conclusion – Prebunking ads improved information literacy, but a greater improvement was shown with AI-generated visualization ads than with audio-only ads. The investigators acknowledge the benefit of theatrical visual advertisements to prebunk misinformation and plan research to include broader populations.