Guided or Misled? AI’s Role in Shaping New Voters

Anna Bulto ’26 and Skylar Li ’26

In early April, a virtual assistant powered by artificial intelligence texted undecided voters during the Wisconsin Supreme Court elections, offering to answer their questions. The model was created by Defend Our Votes — an organization working toward a pro-Democracy majority in the courts — and Convos — an AI company focused on AI SMS conversations for organizations of all kinds. It was trained with the intention of convincing more than 1 million undecided voters to vote for liberal Justice-elect Chris Taylor. This was not the first time AI has been used to support political candidates: In the 2024 presidential election, New Hampshire voters received AI robocalls using Joe Biden’s voice. Since then, AI-generated voice calls have been deemed illegal by the Federal Communications Commission.

As rising college freshmen, we and many of our classmates already or will soon have the right to vote. Being first-time voters means that most of us will be unfamiliar with many aspects of the system, and the endless sources of information on politicians can be overwhelming. This could certainly decrease our enthusiasm to vote. Voter education is evidently extremely important, not only for politicians and parties, but also for the protection of democracy itself. AI models that can have personalized conversations with first-time voters have immense potential, as they have access to unlimited information at an unparalleled speed. This could be extremely useful to help educate new voters, such as ourselves, and increase voter turnout. In this case, the model used to text undecided Wisconsin voters was trained on a smaller data set from Defend Our Votes. Because of this, there was a substantial risk of misinformation and bias. Yet, it is a fact that any data that models are trained on has minor biases. Using a larger language model such as ChatGPT or Gemini for such texts could possibly decrease bias, but not eliminate it. 

With this unavoidable bias in mind, we wonder if using any AI tools should be banned in political applications, as AI-generated voice calls were. Although this would eliminate AI misinformation in this regard, banning these models has parallels with limiting parties’ rights to campaign. Is it different for a human to try to persuade voters through conversations versus an AI model? Where do we draw the line between a tool for persuasion and one for manipulation? One thing that is certain is that AI models used in this manner must fully disclose their intentions, especially when they were developed to support a specific candidate, to prevent a voter from relying on one biased source. This was done by the virtual assistant used in the Wisconsin elections, who clearly stated that it was created by Defend Our Courts and supported Judge Chris Taylor. 

The use of AI has lots of potential in political applications, as it makes information about voting systems and candidate policies much more accessible, which is especially helpful for young first-time voters. Yet, it is crucial that we approach AI-generated political information with great caution and recognize that the models can amplify biases that are already pronounced in the political atmosphere.

Work Cited

Conquest-Ware, Steph. “Thousands of Wisconsin voters conversed with an AI chatbot before the Supreme Court election.” Wisconsin Public Radio, Wisconsin Public Radio, 13 April 2026, https://www.wpr.org/news/wisconsin-voters-ai-chatbot-2026-supreme-court-election. Accessed 20 April 2026.