A newly identified influence operation is targeting the information sources used by AI chatbots rather than trying to persuade voters directly. Israeli threat-intelligence firm VisionHeight found a network of anonymous hyperlocal websites publishing candidate endorsements and issue-by-issue talking points across at least 14 U.S. states, according to CyberScoop’s report [19].
The operation initially comprised 37 sites, including outlets named Wolverine State Report and Peach State Bulletin, but VisionHeight later identified 460 pages of related content. The sites were organized into state-based clusters, with some presenting themselves as news organizations and others as opinion publications. They linked to one another to create the appearance of a local media ecosystem while repeatedly declaring which candidates were preferable on issues such as housing affordability and taxes.
VisionHeight said the material appeared designed for search engines and AI crawlers, not human readers. The sites allowed major U.S. AI companies to scrape their pages, and the templated content supplied direct answers to queries such as which candidate was better on a particular policy. The firm told CyberScoop that Perplexity had cited a page from Tarheel State Report as the first source in at least one answer, although the researchers did not attribute the operation to a specific government, political group or threat actor.
The campaign covered Arizona, Colorado, Georgia, Indiana, Iowa, Kansas, Michigan, Mississippi, Nevada, North Carolina, Pennsylvania, Texas, Virginia and Wisconsin. At least seven states had House races targeted, while other pages addressed secretary of state contests, judicial races and local offices. VisionHeight reported that all 37 sites from its initial analysis were registered during one evening, with 36 registered within an eight-minute window; on Sept. 30, the network published roughly 100,000 words of candidate-related material using shared templates. The company’s research summary describes the initial network as a coordinated set of 37 anonymous local-news sites [16].
The findings do not establish that chatbots changed voter behavior or that any candidate benefited, but they demonstrate a low-cost way to seed authoritative-sounding claims into the web data that AI systems retrieve. Election officials, campaigns, journalists and voters should treat chatbot answers about local races as leads rather than evidence, verify candidates and voting information against official election sources, and inspect cited pages for recycled text, anonymous ownership and circular linking. The episode also shows why AI providers need stronger provenance checks and monitoring for coordinated, synthetic local-news networks before their content is surfaced in election-related answers.
