A West Virginia congressional campaign is using AI texts to turn voter replies into staff summaries and alignment scores, adding a research function to outreach ahead of November’s U.S. midterms.
Democratic congressional candidate Ace Parsi’s campaign says it partnered with startup Convos and supplied messages written by Parsi or campaign staff. The campaign says its district has more than 500,000 voters and that it cannot speak with each one individually. It says it will not share or retain individual data after the election, while keeping high-level findings.
Campaign manager Monica Biddix told CBS News that staff receive each chat with a summary and alignment score. The campaign told CBS News it sent 23,000 texts and about 400 recipients had a back-and-forth conversation, roughly 1.7% by the campaign’s figures. A campaign post separately said 300 people had texted with the software; the reports do not reconcile the counts or establish whether they cover the same period or type of interaction.
Marty Santalucia, a partner at Vector Political, told CBS News his firm sent more than 3 million texts and generated more than 35,000 individual conversations during the 2026 cycle. Convos says its outreach before Wisconsin’s April 2026 Supreme Court race reached more than 1 million voters and produced more than 21,000 two-way conversations.
A Nature study published in December 2025 involved 2,306 U.S. participants in an experiment tied to the 2024 election. The study measured self-reported candidate preferences, not votes. On a 100-point scale, preferences among Trump supporters shifted 3.90 points toward Harris after policy-focused chats; among Harris supporters, they shifted 2.30 points toward Trump. The conversations had three rounds and a median duration of 6.1 minutes. The authors compared the results with political-ad experiments from 2016 and 2020, while cautioning that differences in study design limit direct comparisons.
A separate U.K. study published in Science analyzed 91,000 AI conversations involving nearly 77,000 participants. It found that methods that increased chatbot persuasiveness tended to make its claims less accurate.
Campaigns’ use of AI texting during November’s midterms will provide another chance to observe how they deploy the tools. Whether the exchanges affect voting remains unknown.
