Protecting Voters in the Era of Generative AI

When OpenAI released ChatGPT just after the 2022 midterm elections, it became clear that U.S. democracy had entered a new phase — one where generative artificial intelligence could impact people’s access to voting information. 2024 was the first federal general election cycle of this generative AI era, with large language models (LLMs) making it possible for voting information to reach audiences in new ways.

However, providing reliable voting guidance through this new technology is a complicated process. Elections in the US are managed by state and local election officials, which means voting guidance varies substantially from state to state, county to county, and even election to election. Since registering to vote and casting a ballot often require following complex instructions, inaccurate voting guidance can become misinformation that disenfranchises voters.

There is also a significant threat that bad actors will attempt to mislead and manipulate voters on a massive scale using these powerful new technologies. Disinformation campaigns have already been successfully deployed in recent elections with tools that are far less sophisticated than generative AI.

To mitigate these risks, new LLMs must use additional integrations to ensure the content they generate is accurate and voter engagement organizations must provide their audiences with reliable election information. Democracy Works data can play a critical role in these initiatives.