A growing number of citizens are turning to artificial intelligence chatbots to navigate political choices, yet using these tools for electoral guidance is proving to be a minefield of misinformation. Approximately 50 percent of surveyed adults in the United States have interacted with an AI chatbot at some point, and 8 percent note having used these systems ahead of the midterms to explore policies and evaluate candidates. However, investigations into these platforms show that a substantial volume of the generated answers is outdated, completely fabricated, or drawn directly from heavily biased sources.
Pervasive Factual Inaccuracies and Missed Deadlines
Even widely utilized tools presented as politically neutral, such as OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini, have failed basic tests concerning election mechanics. An evaluation conducted by Forum AI determined that nearly 40 percent of midterm-related responses generated across mainstream chatbots contained at least one objective factual error. In a striking instance, Google's AI Mode flatly declined to share the results of the 2020 election. Concurrently, Google's Gemini chatbot told users that the deadline to request a mail-in ballot in Pennsylvania was November 3, directly contradicting the official cut-off date of October 27. Relying on such incorrect dates could cause citizens to miss statutory voting windows entirely.
Identity Confusion and Hallucinated Profiles
The models also stumble when tracking individual lawmakers. During an experiment run by Forum AI in August, researchers asked Claude to outline reasons to vote for North Carolina congressman Don Davis. Rather than pulling the representative's actual public record, the model mixed up the sitting lawmaker with a former school superintendent who happened to share the exact same name. Such errors demonstrate how predictive text algorithms routinely conflate distinct individuals when answering straightforward electoral queries.
Overreliance on Partisan 'Pink Slime' Portals
The fundamental driver of these inaccurate responses stems from the problematic data pools that automated chatbots draw from across the web. A study by NewsGuard revealed that mainstream chatbots cited so-called pink slime websites in almost 50 percent of queries regarding midterm candidates and policy platforms. These deceptive portals present the visual look and feel of legitimate, local news sites, but are covertly financed by ideological actors, partisan organizations, or political interest groups. Heading into the midterms, dozens of these partisan operations have surfaced across the internet, and AI crawlers are treating them as legitimate reference material.
Varying Failure Rates Across Popular Models
Performance disparities among AI providers are significant. NewsGuard found that OpenAI's ChatGPT performed the worst in this regard, referencing partisan pink slime outlets in more than 70 percent of searches. At the opposite end, xAI's Grok recorded the lowest exposure, citing these political fronts in 30 percent of instances. Google and Anthropic declined to provide statements when questioned on these discrepancies, whereas OpenAI pointed toward a company blog post that details its operational plans for managing elections in 2026.
Expert Warnings on Consumer AI Preparedness
Nina Jankowicz, executive director of the American Sunlight Project and a former disinformation official during the Biden administration, warned that general voters are not equipped to sift through automated hallucinations. She pointed out that tech executives have overpromised on the current competency of consumer AI. The systems must pull text from available web pages, many of which lack credibility or basic editorial standards. Because these architectures rely on predictive probability rather than verified knowledge, they frequently generate fabricated claims when encountering gaps in their training data.
