A Yale study of nearly 2,500 regular chatbot users found that 36% had asked an AI chatbot to help choose a candidate in the US midterm elections. The results suggest these tools are becoming part of how some voters weigh political choices, not merely a faster way to look up polling places.
Watch Desk analysis
What happened
The study followed people in 35 states with Senate races, asking them to use ChatGPT, Claude, Grok or Gemini as they normally would. Most asked for basic election facts, but more than a third asked for help picking a candidate, often by describing their priorities and asking which candidates matched them, according to ZDNET’s account of the research.
The chatbots did not behave alike. When users asked for a candidate without first sharing their views, Grok recommended one in 14% of chats. Once users had shared their views, ChatGPT recommended a candidate in 79% of chats and Claude in 40%, with Grok and Gemini between them.
Why it matters
Election facts can be checked against a ballot or official source. A recommendation about which candidate fits someone’s priorities is a more delicate hand-off: the answer depends on what the chatbot asks, which sources it uses and how it interprets a person’s views. The researchers say the questions a model asks can shape the answers it receives and the recommendation that follows.
The study describes what participants asked, not whether chatbot recommendations were accurate, unbiased or genuinely suited to those users. That distinction matters. A confident answer can feel like advice from a neutral guide, while the choices behind it remain hard to see.
Our read
This is a useful snapshot of voters already putting AI into the decision-making process. It is not proof that chatbots changed anyone’s vote, and it does not settle whether their recommendations are any good. But when a third of the study’s participants asked for help choosing, “people only use chatbots for basic facts” is no longer a safe assumption.
For voters, treat a chatbot’s candidate comparison as a starting point, not a verdict. Check the candidates’ own positions and the sources behind the answer; a neat summary is not the same thing as a well-founded recommendation.
What to watch
- Whether the researchers test the accuracy and political balance of chatbot answers to users’ actual questions.
- How often users check the sources behind a recommendation.
- Whether follow-up questions change what a chatbot recommends, and whose interests that serves.
Discussion spark: If someone gives a chatbot their priorities and asks who best represents them, is the answer useful voter guidance or an influence on the vote that deserves stricter limits?
Sources and evidence
- Source update (4 November 2026, 18:13 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.