Large language models are increasingly used for election information, but little is known about how external web content shapes the political information they present. A new paper on arXiv introduces VoteGEO-Bench, a benchmark designed to measure the impact of generative engine optimization (GEO) on electoral visibility.

The study focuses on GEO, a practice that rewrites web content to influence the outputs of generative engines. VoteGEO-Bench appears to offer a systematic way to assess how such optimization affects which political information becomes visible in LLM responses.

The authors note that this area is understudied, and the benchmark aims to fill that gap. While the abstract does not provide specific results, the contribution is the benchmark itself, which could support future research into the intersection of web content, LLMs, and electoral information.