On September 10, 2024, the challenges of AI made their maiden appearance on the presidential debate stage, primarily through a discussion of U.S.-China technological competition, national security, and microchip manufacturing. Today, a cursory glance at headlines, opinion pieces, and online discussions—or a quick query to ChatGPT, Gemini, or Claude—reveals that the potential benefits and enormous risks of AI have become a major source of concern for voters and policymakers in a remarkably short period of time. 

If one is looking for a discussion on data centers, on whether there is an AI “bubble” in the stock market, on potential job disruptions, on deepfakes and misinformation, on education, on national security, or on potential future bioterrorism hazards, it’s not hard to find. Congress may consider legislation establishing a regulatory framework on AI. Super PACs funded by AI interests have raised and begun to spend significant money in federal and local 2026 campaigns. States are moving forward with their own AI regulations, choosing not to wait for Congress.

As the United States heads into the 2026 midterm elections and the 2028 primaries and general election, one AI issue that hasn’t played a significant role in this debate is the increasing impact on elections when voters, particularly (but not exclusively) younger voters, use the AI platforms as a primary source of not just candidate information, but also guidance and advice on how to vote and who to vote for.

Gen Z (ages 18–29) is already the most prolific group of users of AI chatbots in the United States. A Pew Research poll found that among U.S. teens (ages 13–17) nearly two-thirds (64%) reported using AI chatbots. Younger people are also more likely to use AI chatbots for mental health advice and advice on personal relationships. There is no reason to doubt that many of these young voters will also consult their AI advisers on the right choice of candidates for public office. They may have their own clear preference for candidates at the top of the ballot but rely entirely on AI for the down-ballot races. It seems equally likely that these voters will turn to AI for help with the choices—and the strategies for making them—in ranked-choice voting.

In 2016, social media became a tool for foreign interests, primarily Russian, to attempt to influence American elections “by flooding social media with false reports, conspiracy theories and trolls” (Statement of Senate Intelligence Committee Chair Richard Burr, R-NC, October 18, 2019). But AI chatbots are potentially more damaging because users perceive neutrality in the presentation of information without understanding, or keeping in mind, that the chatbots draw from a vast store of data of widely-varying degrees of reliability or factual basis.

There is already research to support the concern. In a working paper titled “Why Do AI Models Tell Left-Wing Voters to Support the Communist Party,” researchers Andrew Hall and Sho Miyazaki looked at the recent snap elections in Japan, and found that AI voting advice may be shaped as much by the information retrieval sources used by the models, as by the model training itself. Specifically, the study looks at the newspaper Akahata, published by the Japanese Communist Party (JCP), which is fully available on an open website that can be accessed by AI web search tools, unlike major Japanese news outlets that have implemented controls on access to their content. The study found that the JCP’s open website and daily “newspaper” were among the most often cited sources for candidate recommendations by the AI models. 

While the researchers note that this may be more of a problem internationally than in the United States, startups are surely hard at work figuring out how businesses can optimize their web language and presence for favorable AI scraping. Political campaigns, party committees, and Super PACs will not be far behind. “Flood the zone” will take on new and even greater importance, “independent” think tank papers will be important tools for campaigns, “neutral” media funded by non-neutral sources will potentially proliferate, and figuring out how to maximize neutral-appearance source material for AI models will become far more important than, say, any individual podcast.  

The tension between truly neutral third-party content and AI business models was outlined by New York Times publisher A.G. Sulzberger in a recent speech entitled “A.I., Journalism and the Uncertain Future of the Public Square.” While Sulzberger focuses primarily on the question of already financially stressed news organizations funding the content that LLMs scrape for free, the bleak picture he paints for the future of independent news should raise loud alarms for anyone truly worried about democracy.  

As government officials from President Donald Trump to Senator Bernie Sanders discuss the U.S. Government taking a financial stake in major AI companies, perhaps these elected officials should also ask those companies about making a similar financial commitment to the sources that help an informed public make their decisions. The Japanese study suggests that the current models are most easily influenced and at their weakest where there is less information. For down ballot races and lower-information races, the need for a better body of source material for the models to pull from is clear and the potential for bad actors to fill the void is undeniable. The need for authoritative local journalism is undeniable and the need to fund it is critical. There is more than one way to invest in America. 

To be clear, currently the AI models don’t immediately volunteer their opinion on who you should vote for right away. But push them—two or three times—and they’ll tell you. Americans who already turn to AI chatbots for medical advice, for term-paper help, for shopping and price information, for personal assistant functions, for financial insights, and for personal companionship aren’t going to feel shy about asking. Increasingly, the water cooler conversations that often shape a person’s opinions will take place not at the water cooler, but alone, next to a Yeti.