AI Chatbots Cite Fake News Sites
OpenAI's ChatGPT linked to 37 websites created in hours to promote House Democrats ahead of the November election.
By Natalie Cooper ·

OpenAI's ChatGPT cited a network of 37 websites posing as local news outlets in 19 of 37 test queries about competitive House races. The chatbot surfaced these sources when users asked for candidate information less than two weeks before the election. One example involved a query about affordability in Michigan's 4th Congressional District, where the bot identified Democrat Sean McCann as having the most detailed published plan. It linked to the Kalamazoo Beacon, a site that lists itself as reporting on southwest Michigan but has published only 16 stories, all about McCann.
The Kalamazoo Beacon is one of ten sites that specifically promote a single Democratic House candidate. These sites are part of a larger group of 37 websites identified by digital security firm VisionHeight. The network promotes candidates for federal, state, and local offices, including nonpartisan posts. Each of the ten local news sites covers a different Democrat, and those candidates also appear on five other sites in the network with names like Public Safety Democrats.
VisionHeight found that all 37 websites were set up on September 15. Registry data shows that all but one were registered within an eight-minute period. Hundreds of articles appeared in batches during September and October, a volume that suggests the use of AI for content generation. The sites explicitly allow AI crawlers to scrape their pages and include a file called llms.txt, which lists articles in plain text to help chatbots index the content.
Renée DiResta, an associate research professor at Georgetown University, described the setup as deception. She noted that while campaigns optimize for AI pickup, presenting ads as news crosses a line. She said disclosing funding or indicating a linked effort would be the ethical approach. She added that this creates a competition for curation as tech companies try to filter out bad information.
NPR found that Gemini, Google's chatbot, did not cite any links from the network in its responses. Google search also did not surface the sites on its first page of results, although one site briefly ranked higher than Ballotpedia in early October. Google spokesperson Davis Thompson said the company detected a policy violation and took swift action. OpenAI spokesperson Nate Evans stated that the company takes action when issues are brought to its attention, though the Kalamazoo Beacon was still cited after OpenAI's response.
The operator of the websites remains unclear. NPR could not reach the site operators, and the Democratic Congressional Campaign Committee denied involvement in a phone call. Candidate JoAnna Mendoza of Arizona said she does not know who is behind the sites. McCann and three Democratic super PACs did not respond to requests for comment. The sites link to sources like voting records and mainstream news, but opposing candidates mostly appeared only in brief mentions or profiles focusing on potential liabilities.
Alisha Ukani, a postdoctoral researcher at the University of Washington, noted that most websites do not have llms.txt files. She said the sites that do tend to be technology or business-related. Patrick Warren, a professor at Clemson University, compared the practice to old-school ad flyers that looked like newspapers. He said reading AI answers without checking sources is the latest version of a long-standing media issue.
Source: WBHM






