A Yale study surveyed nearly 2,500 U.S. voters across 35 states during the 2024 midterm elections, uncovering that AI chatbots often provide inconsistent election-related assistance, especially when asked for candidate endorsements or issue alignment.
- 36% of users asked chatbots for candidate voting advice, with diverse recommendation patterns.
- ChatGPT provided candidate endorsements most frequently after users shared their views.
- Misinformation and outdated content remain risks with AI election assistance.
What happened
Yale researchers conducted a study with nearly 2,500 adults who regularly use AI chatbots, examining their interactions with ChatGPT, Claude, Grok, and Gemini around the 2024 U.S. midterm elections. Participants primarily requested basic election facts, but over a third sought help in choosing candidates based on their stated priorities. Notably, Grok occasionally recommended candidates proactively, whereas ChatGPT tended to recommend candidates only after users provided personal views.
The study demonstrated significant variation in chatbot behaviors: ChatGPT recommended candidates in 79% of chats after users disclosed views, Claude in 40%, with Grok and Gemini falling in between. Meanwhile, Pew Research data showed general chatbot use by 42% of U.S. adults but only 8% used chatbots specifically for election insights during this cycle, focusing mostly on candidate issue stances and fact-checking rumors.
Why it matters
AI chatbots now frequently stand between voters and election information, influencing the way people receive and interpret political content. While delivering basic election facts is relatively straightforward for AI trained on vast data, providing unbiased candidate recommendations is more complex and affected by the chatbot’s underlying data sources and algorithms. This variability can lead to conflicting or skewed advice, potentially impacting voter decisions.
Given the widespread risks of misinformation, especially with sophisticated AI-generated deepfakes and disinformation, inaccurate or outdated chatbot responses can mislead the electorate. Despite efforts by AI companies like OpenAI and Anthropic to monitor election content through dedicated teams, user tendencies to trust AI responses without verifying sources heighten the potential for cognitive offloading, where voters rely heavily on AI rather than seeking diverse perspectives.
What to watch next
As AI chatbots become even more integrated into information-seeking behavior, researchers and regulators should monitor how these tools influence democratic engagement and voter knowledge. Tracking improvements in chatbot transparency, source citation, and resistance to disinformation will be critical in assessing their public impact.
Future elections will test chatbot adaptability to evolving misinformation landscapes and user needs. Stakeholders must balance innovation in AI assistance with safeguards that reduce risks of biased or faulty guidance, ensuring voters can critically assess AI-provided election information.