Chatbots cite fake news sites to boost Democratic House candidates
Thirty-seven websites registered in a single day on September 15 are influencing AI search results in key midterm races.
By Matthew Turner ·

Thirty-seven websites promoting federal, state, and local candidates were all registered on September 15, with all but one created within an eight-minute window. This digital network includes ten sites that mimic local news outlets to support specific Democratic House candidates in competitive districts.
NPR testing found that ChatGPT frequently cited these sites when voters asked about candidates. In a query regarding affordability in Michigan's 4th Congressional District, the chatbot placed Democrat Sean McCann at the top of its list, citing the Kalamazoo Beacon. That site has published only 16 stories, all concerning McCann, despite describing itself in its source code as a local news outlet.
The network, identified by digital security firm VisionHeight, operates on a scale that suggests a single coordinator. Hundreds of articles were uploaded in batches during September and October, a pattern VisionHeight linked to the use of AI for content generation. The sites are designed for machine readability, featuring files named llms.txt and welcoming AI crawlers that many traditional news organizations block.
Testing revealed that ChatGPT cited the network in 19 of 37 test questions, or roughly half the time. In 17 of those instances, the chatbot specifically referenced one of the ten sites posing as local news. The chatbot also cited the network for every query about five city council candidates in Michigan. By contrast, Google's Gemini chatbot did not cite any links from the group in its responses.
Renée DiResta, a researcher at Georgetown University, described the practice as a deception that crosses a line. She noted that while campaigns optimize for AI visibility, presenting paid or coordinated efforts as neutral reporting is problematic. DiResta said that disclosing funding would be the ethical approach to this emerging form of stealth advertising.
It remains unclear who operates the sites. The Democratic Congressional Campaign Committee denied involvement, and candidate JoAnna Mendoza of Arizona stated she did not know who was behind the websites. Other candidates and supporting super PACs did not respond to requests for comment. Patrick Warren, a professor at Clemson University, compared the tactic to old-style ad flyers disguised as newspapers, noting that reading AI answers without checking sources is a new version of a familiar problem.
OpenAI spokesperson Nate Evans said the company takes action when issues are brought to its attention. However, the Kalamazoo Beacon was still cited in five test queries after NPR flagged the site. Google detected a policy violation on one of the sites and took action, but the sites continue to appear in AI-generated answers for many races.
Source: KYPX






