American Voters Turn to Claude and ChatGPT for Voting Advice: The Case of the 2026 Midterm Elections

In the United States, as the 2026 midterm elections approach, voters are turning to AI chatbots to decide how to cast their votes. According to a survey by the New York Times, tools like Claude from Anthropic, ChatGPT from OpenAI, and Gemini from Google are being utilized by an increasing number of citizens to navigate the dozens of candidates and referendum questions on local ballots, which are often too numerous and complex to analyze independently within the time constraints of the election campaign.

Typically, users photograph their ballots to submit them to the chatbot, asking for guidance on candidates that align most closely with their political values or on the most effective voting strategies in a particular district. In most documented cases, the systems show an initial reluctance to provide direct indications on political issues, having been trained to avoid answers perceived as biased. However, by reformulating their requests - for example, asking the chatbot to rely on candidates' voting histories or on journalistic sources considered authoritative - users often manage to receive operational guidelines, including recommendations for strategic voting to favor certain electoral outcomes.

Anthropic has stated that users posing political questions to Claude should receive complete, accurate, and balanced responses, capable of helping them reach independent conclusions rather than steering them toward a specific position, emphasizing that the model is trained to treat different political orientations with equal depth and analytical rigor.

However, experts consulted by the American newspaper highlight several critical issues. David G. Rand, a professor of information science at Cornell University, points out how the persuasiveness of these models stems from their ability to present information clearly and seemingly factually, while warning that the results tend to reflect the implicit biases in the questions posed by users, rather than offering a truly neutral analysis. Yamil Velez, a political science professor at Columbia University, notes that an ideal tool should draw from a curated and verified database of political programs, rather than gathering information indiscriminately from the web, as is currently the case. The scholar also indicates a potential distorting effect: candidates more present in local media and on social media are more easily accessible by chatbots, and political communication strategies are already adapting to this scenario, favoring structured content - such as bullet points - that is more easily processed by language models.

Despite these concerns, several voters interviewed describe their experience in positive terms, reporting a significant time savings compared to the hours traditionally dedicated to independent research on candidates and a greater confidence in their final choice. Nonetheless, there are forms of caution: some users manually verify the links and sources cited by the chatbot before proceeding to vote, and in general, they avoid communicating their actual voting choices to the AI, limiting themselves to sharing blank ballots.

The phenomenon, which could also affect the Old Continent and Italy in the future, places a crucial node at the center of public debate: the increasing delegation of a cornerstone process of democratic life to generalist artificial intelligence systems, not specifically designed for electoral analysis and lacking structured source verification mechanisms. The fear is that, somehow, voting could be influenced even more than in the past, through the responses of an oracle that seems to have an answer for everything. Whatever that answer may be.