West Virginia Democrat uses AI to text voters ahead of midterms

Political campaigns across the United States are increasingly relying on artificial intelligence to communicate with constituents, a shift that has drawn scrutiny from researchers and privacy advocates. Ace Parsi, a Democrat running for Congress in West Virginia, has adopted this strategy despite describing himself as personally against the technology. His race against Republican Representative Riley Moore in a district that voted heavily for the Republican party in the previous cycle is one of the most competitive in the state. Parsi and his staff decided to use AI chatbots to send text messages to voters before the November election.

The bots are powered by large language models, the same technology behind consumer tools like ChatGPT. Campaigns load specific materials, such as press releases and speeches, into the system to restrict the bot's responses. This prevents the AI from drawing on the open web. The technology allows a single campaign to hold thousands of conversations in one day. A study published in the journal Nature in December found that these short, AI-led dialogues shifted voter opinions more than traditional political advertisements. In experiments involving the 2024 U.S. presidential race, pro-Harris chatbots moved Trump supporters nearly four points on a 100-point support scale. That effect was roughly four times larger than the impact of pro-Democratic video ads in previous election cycles.

However, the technology is not without flaws. A companion study in the journal Science found that when chatbots push harder to persuade, their claims become less accurate. The study tested 19 large language models across more than 91,000 conversations. Most states do not currently require campaigns to disclose that voters are texting with an AI. This has raised privacy concerns, as campaigns can receive full transcripts of these private exchanges. Marty Santalucia, a partner at Vector Political, said his firm has sent over 3 million text messages this cycle, resulting in more than 35,000 individual conversations. He noted that 10% to 20% of respondents who engage with the bot stay in the conversation for at least an hour.

Parsi's campaign manager, Monica Biddix, confirmed that the campaign receives every individual chat, along with summaries and alignment scores for each voter. The system rates individuals on a scale from 0 to 100, where 100 indicates high motivation to support the campaign. Parsi said his team told voters at the start of each conversation that they were speaking with a chatbot. He used the data to call voters personally, specifically those who expressed concern about the tool or ranked low in support. "I actually did the opposite. I mean, the ones that I called a lot were the people that disagreed with me," Parsi said.

Katie Harbath, CEO of Anchor Change, described the technology as still being in an experimental stage for many campaigns. She noted that firms are working to refine how these agents listen and persuade. Jessica Alter, co-founder of Tech for Campaigns, argued that the tool allows campaigns to reach voters who prefer texting to door-to-door canvassing. The next step for Parsi's campaign is to follow up these digital interactions with human phone banking. The midterm elections are scheduled for November 2026.