37 Websites Registered in 8 Minutes Cited by ChatGPT in House Races

A network of 37 websites, most of which were registered within an eight-minute window on September 15, is appearing in AI-generated search results for competitive U.S. House races. The sites, identified by digital security firm VisionHeight, promote a majority of Democratic candidates while mimicking the format of local news outlets. In 19 of 37 test queries, ChatGPT cited at least one article from this network. This occurs less than two weeks before the midterm election, when voters are increasingly relying on chatbots for candidate information.

The most prominent example is the Kalamazoo Beacon, a site that describes itself in its source code as reporting on southwest Michigan. The site has published only 16 stories, all of which feature Democrat Sean McCann, the candidate for Michigan's 4th Congressional District. When asked which House candidates in that district care about affordability, ChatGPT cited the Beacon and stated that McCann has the most detailed published affordability plan. The Beacon labels its content as news articles in its metadata, despite the one-sided coverage.

The 37 sites function as an interconnected ecosystem. Ten of them pose as local news outlets, each covering a single Democrat in a competitive race. The remaining sites include state-level political blogs and roundup-style articles that feature candidates from multiple districts. According to VisionHeight, the volume of content suggests the operator may have used artificial intelligence to generate the material. Hundreds of articles appeared in batches during September and October. The sites also include a file called llms.txt, which lists articles in plain text to facilitate scraping by AI crawlers. Most traditional news websites do not use this file.

Renée DiResta, an associate research professor at Georgetown University who studies online influence campaigns, called the practice a form of deception. She noted that while campaigns optimize for AI pickup, pretending to be a neutral news source crosses a line. DiResta said that disclosing funding or indicating a linked effort would be the ethical approach. She added that opposition research groups are also trying to ensure negative information about candidates is found, creating a competition for curation in the AI search space. The sites are not entirely hidden; they link to congressional voting records and mainstream news articles to support their claims.

The operation appears to have a single operator, though identity remains unclear. Registry data from VisionHeight and Andreas Sjöstedt of Erebus, a threat forensics firm, shows all sites were set up on September 15. All but one were registered within eight minutes. The Democratic Congressional Campaign Committee denied involvement in a phone call. American Bridge 21st Century, Tomorrow Together, and House Majority PAC did not respond to requests for comment. JoAnna Mendoza, a Democrat in Arizona promoted by the network, said she does not know who is behind the sites.

Google's products behaved differently than OpenAI's. In the 37 test queries, Gemini did not cite any links from the network. Google search also did not surface the sites in its first-page results. However, VisionHeight found that in early October, at least one site ranked higher than Ballotpedia on Google's first page. Google spokesperson Davis Thompson said the company detected a policy violation and took swift action. He stated that generative AI features are built on ranking systems designed to block spam and abuse.

OpenAI spokesperson Nate Evans said ChatGPT cites a range of sources so people can make their own judgments. He added that the company takes action when issues are brought to its attention. It is unclear what changes OpenAI has made since the sites were flagged. The Kalamazoo Beacon was still cited as a source in recent queries. In some instances, McCann no longer receives the distinction for having the most detailed affordability plan, but in others, he still does. Patrick Warren, a professor at Clemson University, noted that thinly veiled advertising has a long history, from newspaper-style flyers to search engine ads. He said reading AI answers without evaluating sources is the latest version of this pattern.

The sites explicitly allow AI crawlers to scrape their content. Many articles are formatted as questions and answers, a structure that chatbots tend to cite. This contrasts with many news organizations that block such crawlers to protect content. The network's success in AI search results may be higher for downballot races. NPR identified five Michigan city council candidates covered by a site called the Wolverine State Report. When asked about their positions, ChatGPT cited that website every time. Preliminary research indicates chatbots cite questionable sources more often when there is little credible reporting, a phenomenon known as a data void.