How Leading AI Chatbots Handle Election Questions Ahead of the US Midterms

Why the policies matter

In a post on Princeton's Freedom to Tinker blog, authors Mihir Kshirsagar and Tanvi Prem point to growing evidence that Americans turn to chatbots for election information. A 2026 Pew Research Center survey found that 42% of US adults now use AI chatbots to search for information. A 2024 Bipartisan Policy Center survey found that 27% of voters were asking AI tools questions about elections at least monthly.

A more recent Verasight survey from April shows a more mixed picture. A majority of adults are skeptical about using chatbots for political information, although at least a third appear comfortable asking about such issues.

The authors distinguish four types of election queries: procedural questions (when, where and how to vote), results (live returns and the status of a race), candidate and ballot information, and the generation of campaign or government-sponsored content such as synthetic audio and video. They sort company responses into four buckets: integrating factual information from a trusted source, responding with neutral information, redirecting to an external source, or refusing the query, sometimes with a banner. They stress that they have not tested how effective these policies are.

OpenAI, Anthropic and Google

OpenAI is partnering with Democracy Works and the Associated Press (AP) this year to provide information on voting processes, locations and live results. The authors describe this as a comprehensive update to its 2024 policies, which focused mainly on preventing deepfakes and pointing voters to CanIVote.org and the National Association of Secretaries of State (NASS). OpenAI bars political ads and scaled campaign content, as well as microtargeting and automated distribution, but supports human-directed political campaign work. It also says it is working with NASS and the National Association of State Election Directors on election execution and cyber infrastructure through Daybreak, and is giving voting system manufacturers and election administrators access to Codex Security and its Trusted Access for Cyber program.

Anthropic shows an election banner in Claude that directs users to TurboVote, a Democracy Works resource. Its 2026 plan expands on the 2024 policy with more detail on preventing bias, evaluating safeguards, and triggering web search for questions about live results. Claude is trained to avoid unsolicited political opinions and to pursue what Anthropic calls "political even-handedness". Anthropic prohibits using Claude for automated campaign messaging, personalized targeting, synthetic media, vote solicitation and disrupting election processes.

Google generally restricted Gemini's answers to election questions in 2024, but has since modified that approach and now allows responses. It offers a Civic Information API for local election information, plans to provide results through AP, and directs Gemini to Google Search, using tools such as SpamBrain to defend against low-quality content. The authors say Google mainly redirects users to external sources rather than integrating facts or giving a neutral response. Advertisers must disclose realistic synthetic content and complete US Election Ad Verification. Google also offers cybersecurity help through its Advanced Protection Program and Project Shield, which protects election websites from DDoS attacks.

Meta, xAI and Microsoft

The picture is thinner for the other three companies. According to the authors, Meta has not published an election-specific policy for Muse Spark or Meta AI. Its 2026 election policies cover only Instagram, Facebook and WhatsApp, and mention C2PA watermarks on AI-generated content, an Elections Operations Center, and directing election searches to government websites.

X has not published an AI-elections policy for Grok or xAI. Since 2024, Grok has started its responses to election questions with a link to Vote.org. X's September 2025 election integrity policy is enforced through Community Notes and civic integrity labels, which X says are reserved for "critical escalations". The authors say it is unclear whether Grok's own outputs fall under that policy. The Grok 4.6 model card includes safeguards against content that creates severe harm or involves criminal activity.

Microsoft has not published a standalone AI-elections policy for Copilot either, though it has discussed voter education and cybersecurity in light of AI. In 2023 it introduced C2PA into its products and a "Campaign Success Team" to help campaigns navigate AI use.

What it means for voters

For voters, the main takeaway is that the biggest providers are moving from restricting or deflecting election questions toward answering them, often with a trusted partner such as Democracy Works or AP behind the answer. Coverage is uneven, however: some companies have detailed rules, while others leave open whether their chatbots are covered by existing election policies at all. Because the authors have not tested how well any of these safeguards work, the policies describe intent rather than proven results.