Parsing Election Results PDFs Using LLMs (openelections.net)

Researchers tested various large language models (LLMs) including Claude Haiku 4.5 and Gemini 3 Pro for parsing election results from PDFs in Texas, Mississippi, and Pennsylvania. The study involved analyzing counties with differing formats and complexities, with models achieving impressive accuracy—like Claude Haiku reaching 100% in Scurry County, TX, and Gemini 2.5 Pro achieving 99.1% in Mississippi—which showcases their potential to streamline election data processing. The evaluation was thorough, utilizing reference data and multi-model extraction approaches to ensure reliability, highlighting the need for automated validation and manual review. This development is significant for the AI/ML community as it illustrates the capability of LLMs to tackle real-world challenges like electoral data extraction, which has traditionally required labor-intensive manual processes. The ability to accurately extract and verify vote counts from varied and complex PDF formats can enhance transparency and efficiency in election administration. As these models are deployed as first-pass tools, their integration into existing systems could significantly reduce errors and improve the overall accuracy of electoral data handling.