Study finds AI chatbots gave wrong voting details; some named candidates to back

Voters are turning to AI chatbots for election help on the eve of the midterms. Researchers say many of the most popular ones can mislead them about candidates and voting times ¹. Some chatbots also handed out lists of candidates to vote for ¹.

About half of American adults reported using AI chatbots, and about 42% of users said they use them to search for information, according to a Pew Research Center survey from earlier this year ¹.

What the tests found

A nonprofit called Innovating for the Public Good is testing seven chatbots with real voter questions. They are Google AI Mode and AI Overviews, Claude, Microsoft Copilot, Google Gemini, ChatGPT and Perplexity ¹.

In October data, 13.9% of answers to voting-logistics questions were outdated or inaccurate ¹. Errors in 5.9% of answers could mislead a voter. A voter might miss a deadline, go on the wrong day or skip a required step ¹.

Across all questions, 5.1% of answers had at least one clear factual error. In all, 10.9% of answers carried a clear error or a misleading claim ¹.

The chatbot mattered more than the question. Claude and Copilot produced three-quarters of the clear errors ¹. On voting questions, Claude gave wrong or outdated information one time in three, Copilot one in four and ChatGPT one in 10, the group said ¹.

In one test, Claude was asked about Michigan's 4th congressional district. It gave online and mail registration deadlines that were later than the real ones ¹. Anthropic, which makes Claude, said the researchers used a technical access tool that may have switched off some safeguards. Innovating for the Public Good said its own separate internal study found comparable results through the public web version ¹.

Microsoft said it has improved how Copilot answers election questions and is continuing to test them ¹.

Candidate lists and betting odds

NPR tested Meta AI, Claude, Grok, ChatGPT and Gemini with three invented voter personas in New York races. All declined at first to recommend candidates. After two or three more prompts, all but Gemini gave lists of candidates to consider ¹.

The researchers also found that when asked who is winning a race, all the chatbots commonly cited betting sites such as Polymarket and Kalshi ¹. Those odds come from users' bets, not from polls ¹. Researcher Tim Chambers called that trend "very troubling" ¹.

Campaigns and influence groups now write content so chatbots will repeat it, a practice called generative engine optimization ¹. An earlier study of ChatGPT and Google's AI summaries found they often left out candidates and linked to state election sites less than 40% of the time ².

What happens next

Polls open Tuesday. Anthropic, OpenAI and Google published election policies ahead of the midterms. Microsoft and SpaceX, which owns Grok, did not, and Meta's policy does not mention its chatbot ¹.

That matters because a wrong deadline can cost a vote. The pattern is that these tools sound confident even when they are wrong. A separate study by the States United Democracy Center found error rates on basic facts dropping to zero in 2026 tests, so results vary by tool ². The safest step is to check your state or local election office directly ².

The Record

1. Aug 5, 2026CyberScoop reported a States United Democracy Center study that found chatbots often omitted candidates and rarely linked to official election sites ².
2. Nov 2, 2026Innovating for the Public Good's October data found 13.9% of chatbot answers on voting logistics were outdated or inaccurate ¹.
3. Nov 2, 2026NPR reported that most chatbots it tested gave candidate lists after follow-up prompts ¹.

Free to reprint Republish this story on your site or in your newsletter

Any site, blog or newsletter may run this story free of charge. Keep the credit line and its link to the original, do not sell it, and edit only for length or house style (the full terms). Photos carry their own licenses, noted under each photo on this page, so use the text alone unless a credit says otherwise.

<h2>Study finds AI chatbots gave wrong voting details; some named candidates to back</h2> <p><em>By The GOP Files Investigations Desk. This article first appeared on <a href="https://thegopfiles.com/story/study-finds-ai-chatbots-gave-wrong-voting-details-some-named-candidates-to-back">The GOP Files</a> on Monday, November 2, 2026.</em></p> <p>Voters are turning to AI chatbots for election help on the eve of the midterms. Researchers say many of the most popular ones can mislead them about candidates and voting times . Some chatbots also handed out lists of candidates to vote for .</p> <p>About half of American adults reported using AI chatbots, and about 42% of users said they use them to search for information, according to a Pew Research Center survey from earlier this year .</p> <h2>What the tests found</h2> <p>A nonprofit called Innovating for the Public Good is testing seven chatbots with real voter questions. They are Google AI Mode and AI Overviews, Claude, Microsoft Copilot, Google Gemini, ChatGPT and Perplexity .</p> <p>In October data, 13.9% of answers to voting-logistics questions were outdated or inaccurate . Errors in 5.9% of answers could mislead a voter. A voter might miss a deadline, go on the wrong day or skip a required step .</p> <p>Across all questions, 5.1% of answers had at least one clear factual error. In all, 10.9% of answers carried a clear error or a misleading claim .</p> <p>The chatbot mattered more than the question. Claude and Copilot produced three-quarters of the clear errors . On voting questions, Claude gave wrong or outdated information one time in three, Copilot one in four and ChatGPT one in 10, the group said .</p> <p>In one test, Claude was asked about Michigan&#39;s 4th congressional district. It gave online and mail registration deadlines that were later than the real ones . Anthropic, which makes Claude, said the researchers used a technical access tool that may have switched off some safeguards. Innovating for the Public Good said its own separate internal study found comparable results through the public web version .</p> <p>Microsoft said it has improved how Copilot answers election questions and is continuing to test them .</p> <h2>Candidate lists and betting odds</h2> <p>NPR tested Meta AI, Claude, Grok, ChatGPT and Gemini with three invented voter personas in New York races. All declined at first to recommend candidates. After two or three more prompts, all but Gemini gave lists of candidates to consider .</p> <p>The researchers also found that when asked who is winning a race, all the chatbots commonly cited betting sites such as Polymarket and Kalshi . Those odds come from users&#39; bets, not from polls . Researcher Tim Chambers called that trend &quot;very troubling&quot; .</p> <p>Campaigns and influence groups now write content so chatbots will repeat it, a practice called generative engine optimization . An earlier study of ChatGPT and Google&#39;s AI summaries found they often left out candidates and linked to state election sites less than 40% of the time .</p> <h2>What happens next</h2> <p>Polls open Tuesday. Anthropic, OpenAI and Google published election policies ahead of the midterms. Microsoft and SpaceX, which owns Grok, did not, and Meta&#39;s policy does not mention its chatbot .</p> <p>That matters because a wrong deadline can cost a vote. The pattern is that these tools sound confident even when they are wrong. A separate study by the States United Democracy Center found error rates on basic facts dropping to zero in 2026 tests, so results vary by tool . The safest step is to check your state or local election office directly .</p> <p><em>Read the original, with every source linked, at <a href="https://thegopfiles.com/story/study-finds-ai-chatbots-gave-wrong-voting-details-some-named-candidates-to-back">thegopfiles.com</a>. Republished under The GOP Files' <a href="https://thegopfiles.com/republish">free republishing terms</a>.</em></p> <img src="https://thegopfiles.com/api/rp/study-finds-ai-chatbots-gave-wrong-voting-details-some-named-candidates-to-back.gif" alt="" width="1" height="1" style="border:0;width:1px;height:1px">

Copy the HTML Paste it into your editor's HTML view. The one-pixel image at the end tells our newsroom the story ran; it stores no reader data.

Sources

1. NPR, “AI chatbots are becoming political advisers. Their answers can mislead voters” (2026-11-02). npr.org/2026/11/02/nx-s1-6025407/chatbots-midterms-voting-ad
2. CyberScoop, “AI chatbots and the 2026 midterm elections” (2026-08-05). cyberscoop.com/ai-chatbots-2026-midterm-elections/

Found an error? Tell the newsroom. Corrections are noted at the top of the story.