Before Election Day, some voters have begun using AI chatbots to help make voting decisions. An NPR investigation revealed that OpenAI’s ChatGPT suggests websites posing as local news outlets, which are actually political advertisements.
ChatGPT’s Role in Elections
In a competitive U.S. House race in Michigan’s 4th Congressional District, ChatGPT praised Democrat Sean McCann for his affordability plan. This response could influence voter opinion in races crucial to controlling the House.
When NPR inquired about McCann, ChatGPT cited the Kalamazoo Beacon, a site labeled as a news outlet. However, this site only published articles favorable to McCann, highlighting the deceptive nature of some sources AI refers to.
Analyst Renée DiResta notes that political campaigns aim to exploit AI chatbots for influence, which blurs ethical lines. These misleading citations by AI can affect voters, who increasingly rely on these tools for information.
Websites and Political Deception
The Beacon is part of a network of 37 sites promoting various candidates, mostly Democrats. VisionHeight identified these sites, showing the deceptive appearance of news while promoting political agendas.
Efforts to contact those behind these websites were unsuccessful. Candidates like JoAnna Mendoza and the Democratic Congressional Campaign Committee denied involvement, highlighting the obscure nature of the operation.
Impact of the Network
Misinformation from these sites can skew perceptions of candidates, especially in lesser-covered races. The power of these AI-recommended sources illustrates the ongoing challenge in combating deceptive practices in online information.
The 37 sites developed rapidly, likely using AI to generate content quickly. All were registered on the same day in September, suggesting a coordinated effort to influence political messaging.
Chatbot and Search Engine Responses
ChatGPT alone often cited these sketchy sources alongside reliable ones like Ballotpedia. Questions about specific candidates frequently led to these deceptive websites.
In contrast, Google’s AI tools and Gemini avoided these sources, showing variations in how AI systems validate content. Google quickly removed any instances where these questionable sites appeared in search results.
OpenAI acknowledged the issue but did not specify changes made to rectify the situation after NPR’s findings. Instances with McCann as an example showed inconsistent messaging.
Historical Context and Methodology
Deceptive advertising disguised as news isn’t new. This practice continues in the digital age with AI, adding complexity to evaluating political messages.
NPR utilized comprehensive analysis, downloading pages and examining changes over time. They manually reviewed network articles and used AI for deeper insights, confirming the biased portrayal of opposing candidates in this context.
Carolyn Hax Advises on Managing Relationship Resentment 
Challenges in Polling the Kansas Senate Race 
AI Chatbots and Their Impact on Youth 
Gov. JB Pritzker’s Path to a Third Term 
Florida’s Amendment 3: A Complex Decision for Voters 
Senate Race Overview: Competitive States and Election Prospects