Policy

Mainstream AI Chatbots Fail Election Fact-Checking, Forum AI Research Shows

Study finds 40% of election answers from ChatGPT, Claude, and Gemini contain factual errors ahead of 2026 midterms.

Last verified:

Election AI Accuracy Crisis

Mainstream AI chatbots are supplying voters with systematically unreliable election information, according to research released by Forum AI and reported by Wired. In testing across OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini, Forum AI found that nearly 40% of responses to election-related queries contained at least one factual error. The errors ranged from outdated mail-in ballot deadlines—Gemini incorrectly stated Pennsylvania’s deadline as November 3 when the actual cutoff was October 27—to identity confusion, such as Claude conflating North Carolina congressman Don Davis with a former school superintendent bearing the same name.

According to Wired, these failures occur despite the platforms’ non-partisan positioning and occur at scale. Approximately 8% of recently surveyed US adults have used AI chatbots to research candidates and policies, according to Pew Research data cited by Wired, making the accuracy problem a direct threat to electoral decision-making.

The “Pink Slime” Problem

The root cause lies in contaminated source material. According to NewsGuard research reported by Wired, mainstream chatbots cite so-called “pink slime” websites—sites that mimic legitimate news outlets but are funded by political or ideological groups—at a rate of nearly 50% when answering questions about candidates and midterm policies. The Wired article reports that ChatGPT was identified as the worst performer among tested platforms in this category, while Grok performed comparatively better.

Wired notes that the proliferation of these partisan-funded sites ahead of the midterms has accelerated the problem, giving chatbots an increasingly degraded information diet to draw from.

Hallucination and Training Data Decay

According to Wired, Nina Jankowicz, executive director of the American Sunlight Project and former disinformation czar during the Biden administration, characterized the core vulnerability: chatbots are “predictive” systems that “frequently make things up” when they lack reliable source material. The information they ingest is often outdated—training datasets have knowledge cutoffs—or sourced from low-reliability origins, leaving them structurally incapable of providing authoritative, real-time election guidance.

Wired reports that Google and Anthropic did not respond to requests for comment, while OpenAI directed the publication to a blog outlining the company’s election-focused approach for 2026.

Why This Matters

Voters facing a 2026 election cycle are increasingly likely to encounter AI-generated election information, yet no major chatbot has demonstrated reliable accuracy on ballot-specific details or candidate identification. The near-40% error rate documented by Forum AI means that nearly half of users consulting these platforms for voting guidance will encounter false information—often presented with high confidence. For election officials and platforms concerned with electoral integrity, this research underscores the urgent need for chatbot developers to either implement real-time fact-checking pipelines, clearly label election-related content as unreliable, or restrict generative responses on time-sensitive voting matters entirely. Until those safeguards are in place, the Wired reporting suggests voters should treat AI chatbots as a starting point for research, not an authoritative source.

Frequently Asked Questions

How reliable are AI chatbots for election information?

According to Forum AI research cited by Wired, mainstream chatbots including ChatGPT, Claude, and Gemini produced factually incorrect responses in nearly 40% of election-related queries tested, making them unreliable sources for voting decisions.

What is a 'pink slime' website?

According to NewsGuard's research reported by Wired, 'pink slime' sites present as mainstream news outlets but are actually funded by political or ideological groups and individuals, creating the appearance of legitimate journalism.

Why do chatbots make errors about elections?

Wired reports that chatbots rely on outdated training data and an increasingly contaminated pool of online sources, including partisan websites. When uncertain, they generate plausible-sounding but false responses—a behavior known as hallucination.

How many Americans rely on AI for election guidance?

According to Pew Research data cited by Wired, approximately 8% of recently surveyed US adults report using AI chatbots to research candidates and policies ahead of the midterms.

#election-integrity #ai-hallucination #fact-checking #chatbots #misinformation