Report: Phony News Sites Manipulate ChatGPT to Aid Democrats
A digital security firm uncovered 37 websites posing as local news outlets that appear to be influencing AI chatbot responses.
A network of 37 websites, disguised as local news sources and political blogs, has been identified as promoting Democratic candidates and potentially influencing user responses generated by the AI chatbot ChatGPT. The findings, detailed in a report by digital security firm VisionHeight, raise concerns about sophisticated methods of voter manipulation ahead of the midterm elections.
The phony websites were found to be cited in ChatGPT's answers to user queries about congressional candidates and their stances on issues such as affordability and public safety. This tactic could misinform voters by presenting biased content as neutral reporting.
One instance highlighted in the report involved the "Kalamazoo Beacon," which appeared to be a fabricated site solely dedicated to promoting Democratic candidate Sean McCann in Michigan's 4th congressional district. According to NPR, every story published on the "Beacon" focused on McCann, and ChatGPT cited it when asked about candidates who "care about affordability."
A spokesperson for the McCann campaign stated that they have no knowledge of or involvement with these websites.
The network of 37 sites includes 10 local news outlets, 10 local opinion blogs, 12 statewide websites, and five national political outlets. All of them appear to advocate for predominantly Democratic candidates, according to VisionHeight.
These sites focus on specific candidates in competitive congressional races across the country, feeding information into AI systems that may then disseminate it to users. NPR reported that ChatGPT cited the network's websites approximately half the time it was queried about one of the targeted candidates.
Evidence suggests a common operator behind all 37 websites. Domain registration data reveals that 36 of the websites were registered within an eight-minute span on September 15, a pattern that, according to VisionHeight and cybersecurity firm Erebus co-founder Andreas Sjöstedt, points to sites likely generated by artificial intelligence for AI consumption.
Hundreds of thousands of words of candidate-related content were published across the network in September and October, often appearing in waves. On a single day, September 30, approximately 100,000 words were posted using a consistent template across various layouts.
VisionHeight observed that favored candidates received lengthy profiles and voter guides containing numerous questions and answers. The phrase "is the stronger choice" was detected over 360 times across the network, an indicator of targeted manipulation aimed at AI bots.
While Google's Gemini chatbot did not cite these websites in testing, and Google search did not prominently feature them, an OpenAI spokesperson stated that the company takes action when issues are brought to its attention. "ChatGPT’s search tools cite from a range of sources so people can explore and make their own judgements," the spokesperson said, adding that integrations with Democracy Works provide direct, up-to-date voting information. The Democratic Congressional Campaign Committee has denied any involvement, according to NPR.
