Using Generative AI for Identifying Electoral Irregularities on Social Media
Sidney Moura, Edney Santos, Pablo Azevedo Sampaio, Kellyton dos Santos Brito · Conference on Digital Government Research · 2025
The increasing use of social media (SM) in political campaigns has raised concerns about electoral irregularities, such as unauthorized voter solicitation and the use of sound trucks. Monitoring and enforcing electoral regulations manually are time-consuming and prone to inconsistencies, highlighting the need for automated solutions. A major challenge in automating the detection of electoral violations is the lack of sufficient labeled data. Additionally, the effectiveness of Generative AI in addressing this issue remains underexplored, especially regarding its ability to create synthetic data and enhance detection accuracy. In this context, this study aims to assess the potential of Generative AI in identifying electoral irregularities on SM, focusing on two common violations in the 2024 Brazilian municipal elections: voter solicitation and the use of sound trucks. The goal is to evaluate whether synthetic image generation, combined with AI-based visual analysis, can improve the identification of such infractions. We first generate synthetic images using Imagen 3, Stable Diffusion, and FLUX, identifying Imagen 3 as the most effective in producing realistic and visually coherent images. Then, we test the ability of three AI models—Gemini 2.0 Flash, Llama 3.2 Vision, and PaliGemma 2—to detect electoral violations in both real and synthetic images. To enhance detection accuracy, we apply different prompting strategies, including basic, chain-of-thought, and detailed prompts. Our findings show that Gemini 2.0 Flash performs best, particularly when using detailed prompts. Also, synthetic images help mitigate data scarcity, improving model training and evaluation. Overall, the study demonstrates that Generative AI, combined with optimized prompt engineering, can significantly enhance the accuracy of detecting electoral irregularities.