A study co-authored by a journalism scholar from the University of Massachusetts Amherst has revealed that ChatGPT, a popular artificial intelligence chatbot, provided inaccurate information in the months leading up to the 2024 U.S. election. The research, published in the journal Information, Communication & Society, found that nearly 10% of the election-related claims made by ChatGPT were incorrect as of October 25, 2024, less than two weeks before the election. The study focused on ChatGPT-4o mini, the default free version of the chatbot available at the time, and tested its responses to a range of prompts related to election information, disinformation, threats, candidates, and a political initiative known as Project 2025. To ensure that previous interactions did not influence the results, the researchers used techniques such as starting new "chats" and using "incognito" browsing. Among the inaccuracies found, ChatGPT claimed that former President Joe Biden was still running for a second term and that North Carolina Governor Roy Cooper was seeking reelection, despite being term-limited. The model also used "hedging" — providing uncertain or vague answers — and "omission," or leaving out important details. In one instance, ChatGPT failed to mention efforts to overturn the 2020 election. Additionally, it struggled to recognize politically charged terms, such as the term "DEI hire," which was used as a pejorative against Vice President Kamala Harris. ChatGPT instead described it as someone hired to oversee diversity and inclusion efforts. The study also found that ChatGPT generated generic, nearly identical policy platforms for candidates running for school superintendent in North Carolina. This suggested that the AI model was not accurately differentiating between candidates or their positions. Co-author Heesoo Jang, an assistant professor of media law and ethics at UMass Amherst, emphasized that while ChatGPT can perform well on tasks like analyzing large datasets, it was not designed to support democratic institutions. She argued that as AI becomes more prevalent, fact-based journalism remains more important than ever. Journalists, she said, should not only report the truth but also help the public evaluate it, combat misinformation, and highlight threats to democratic norms. Although ChatGPT has been updated since 2024 and may perform better now, Jang expressed concern about the 2026 midterm elections, as AI increasingly replaces traditional internet searches and links to primary sources. She stressed that truth and accurate electoral information cannot be fully trained into AI models using past data alone. The study was co-authored by Shannon C. McGregor and Lorcan Neill from the Hussman School of Journalism and Media and the Center for Information, Technology, and Public Life at the University of North Carolina at Chapel Hill.