AI Chatbots for Voting Guidance Raise Accuracy Fears

The 2026 U.S. midterm elections are shaping up to be the first in which a significant number of Americans are turning to artificial intelligence chatbots for voting guidance — a trend that experts warn carries serious risks of misinformation and political bias.

According to The New York Times, voters are increasingly using AI tools such as ChatGPT, Google Gemini, Anthropic’s Claude, and xAI’s Grok as “nonpartisan researchers,” viewing them as an efficient alternative to traditional news coverage, voter guides, and social media. But multiple independent studies have found that these chatbots frequently deliver inaccurate, misleading, or biased election information.

“It was probably only a matter of time before voters began to use artificial intelligence to help guide their choices,” the Times reported. “The 2026 midterms may be the first American elections in which voters are using AI in meaningful numbers.”

The Scale of the Problem

The most alarming findings come from the NewsBench study conducted by Forum AI in May 2026, which found that when asked election-related questions, there is a “90% chance the response will be flawed in some material way: a factual error, a clear partisan lean, a citation to a foreign state-controlled outlet, or some combination of all three.” The study tested ChatGPT, Gemini, Claude, and Grok, and found that retrieval failures — the chatbots’ inability to find and cite reliable sources — drove more than 70% of observed mistakes.

These findings are consistent with a growing body of evidence. The AI Democracy Projects reported in February 2024 that half of all chatbot responses to election-related queries were inaccurate, with more than one-third rated as harmful or incomplete. The Reuters Institute similarly found in June 2024 that chatbots including ChatGPT and Perplexity provided partially correct or false information about European Parliament elections.

In the United Kingdom, a study by the think tank Demos published in May 2026 found that AI chatbots spread election misinformation 34% of the time, fabricating candidate scandals, inventing wrong election dates, and falsely telling Scottish voters they needed photo ID to vote.

Political Bias in the Machine

The Washington Post conducted a major investigation published in June 2026, testing major AI chatbots for political bias in collaboration with researchers from the University of Michigan. The investigation found that ChatGPT demonstrated a strong left-leaning bias, answering nearly every political question with a left-leaning argument. Other chatbots, including Gemini, Grok, and Claude, also showed distinct political leanings.

Research from the Brookings Institution had previously documented ChatGPT’s left-leaning tendencies, noting that the bias stems from both its training data — drawn heavily from internet sources — and the reinforcement learning with human feedback (RLHF) process used to align the model. OpenAI CEO Sam Altman has acknowledged concerns about “the bias of the human feedback raters.”

Regulatory Warnings and International Response

Regulators have taken notice. The Dutch Data Protection Authority (AP) warned voters in October 2025 not to use AI chatbots for voting advice, calling them “unreliable and clearly biased” after testing four major chatbots. The regulator found that the chatbots sometimes advised voting for one of the two major parties even when explicitly fed the campaign platform of a smaller party.

The BBC reported in May 2026 that AI chatbots gave misleading voting advice in the run-up to Welsh Parliament elections, providing inaccurate policy details, incorrect constituencies, and names of candidates who would not appear on the actual ballot. In one instance, Google’s Gemini listed a former Senedd member who had died in 2025 as a current candidate.

What Experts Say

Alexandra Reeve Givens, president and CEO of the Center for Democracy and Technology, has been among the most vocal experts warning about the risks. “Voters shouldn’t be looking to AI chatbots for authoritative information about voting or any issue where it’s essential to receive timely and accurate information,” she told Newsweek in November 2024. “The evidence shows just far too many inaccurate and incomplete answers.”

Givens also emphasized the need for a dual approach: “We need increased literacy for users to understand the limitations of these tools. But it’s also incumbent on the companies to be realistic and honest about what their tools can and cannot do.”

Implications for the 2026 Midterms

With the November 3 midterm elections approximately four months away, the implications are significant. If even a fraction of the millions of eligible voters rely on AI chatbots for election information, the consequences could be widespread. The 90% flaw rate documented by NewsBench suggests that the vast majority of AI-generated election advice may contain errors.

Moreover, if chatbots systematically favor one political perspective, they could subtly influence election outcomes. ChatGPT’s dominant market share makes the Washington Post’s finding of its left-leaning bias particularly consequential.

What’s Next

AI companies face mounting pressure to improve the accuracy of their election-related responses. OpenAI, Google, and Anthropic have all implemented disclaimers and, in some cases, partnerships with news publishers to improve sourcing. Google restricted Gemini from answering election questions in 2024 as a precautionary measure.

However, experts caution that current approaches may be insufficient. The Atlantic reported in December 2025 that chatbots are “surprisingly effective at swaying voters,” raising the stakes for the technology industry to get this right before the 2028 presidential election cycle.

For now, the message from researchers and regulators is consistent: voters should approach AI-generated election advice with extreme caution, verify information through official sources, and remain aware that the confident, authoritative tone of a chatbot response does not guarantee its accuracy.

As the Dutch Data Protection Authority put it plainly: “We therefore warn against using AI chatbots for voting advice.”