Yale researchers reported that 36% of users in a recent study asked AI chatbots to help them pick a political candidate.reported The study involved nearly 2,500 adults in 35 states with Senate races.A new study shows many voters use AI chatbots to help choose which political candidate to support.Yale researchers identified significant variance in how generative models handle subjective political alignment queries versus factual retrieval. Participants were selected because they already use chatbots weekly. This finding highlights how often voters rely on generative AI for subjective political decisions rather than just factual election data.
Evidence: reported
Researchers asked people to use AI chatbots for election info – results were mixedaifleet-zdnet
The figures
ZDNet reported that 36% of study participants asked a chatbot to help pick a candidate.reported When users did not share their views, Grok recommended a candidate in 14% of chats. After users shared their views, ChatGPT recommended a candidate in 79% of chats. Claude recommended a candidate in 40% of those chats. Grok and Gemini fell between those percentages. Pew Research found that 42% of US adults use chatbots to search for information.reported
Evidence: reported
Researchers asked people to use AI chatbots for election info – results were mixedaifleet-zdnet
Evidence: reported
Researchers asked people to use AI chatbots for election info – results were mixedaifleet-zdnet
Behind the figures
The Yale study ran for a few weeks in October leading up to the November 3 midterm elections.reported Researchers asked nearly 2,500 people in 35 states to query ChatGPT, Claude, Grok, or Gemini.reported Participants were adults who already use chatbots weekly. Most users asked for basic facts, such as who was on the ballot. However, many described their priorities and asked the chatbot to assess which candidates aligned with them.
Evidence: reported
Researchers asked people to use AI chatbots for election info – results were mixedaifleet-zdnet
Evidence: reported
Researchers asked people to use AI chatbots for election info – results were mixedaifleet-zdnet
What it means for your stack
Chatbots can skew election-related information and act as an intervening layer in a voter’s search. AI systems can hallucinate and add layers of bias to political information. Interpreting candidate positions is more dicey than retrieving basic facts. It relies significantly on the sources a chatbot uses. A statement from a candidate’s website is more reliable than a social media clip characterizing them.
Building on this
Developers can build interfaces that separate factual election data from subjective candidate assessments. The study shows users often describe priorities to get alignment assessments. Engineers can design prompts that force models to cite source reliability, such as distinguishing between candidate statements and opponent clips. Since Grok recommends candidates without user input, developers might implement guardrails that require explicit user preferences before generating political recommendations. This reduces the risk of unsolicited bias injection.
Why it matters
- Engineers must account for model-specific behavior where Grok recommends candidates without user input, requiring distinct safety checks per model.
- The high rate of ChatGPT recommendations after user input suggests developers need robust logging for subjective political queries to audit bias sources.
- Since chatbots act as an intervening layer, inference engineers should design systems that transparently display source reliability, such as website statements versus social media clips.
- With 36% of users seeking candidate picks, developers face pressure to handle subjective political alignment requests, which carry higher hallucination risks than factual ballot queries.
Topics: generative AI, election integrity, model evaluation, Yale research, chatbot safety
Updated
Then / now / next
Most users asked chatbots for basic election facts like ballot candidates.
Yale researchers found 36% of users ask chatbots to help pick a candidate.
Developers may implement guardrails to require user preferences before political recommendations.
Story timeline
- Then Most users asked chatbots for basic election facts like ballot candidates.
- 04 Nov 2026 Researchers asked people to use AI chatbots for election info – results were mixed
- Now Yale researchers found 36% of users ask chatbots to help pick a candidate.
- Next Developers may implement guardrails to require user preferences before political recommendations.
The strongest doubt
Counterpoint from the sources
The study notes that basic election facts have less room for error due to abundant internet data, whereas interpreting candidate positions is more dicey.
Numbers in context
Circle area is proportional to the magnitude; labels retain signs. Figures may describe different measures. Read their contexts above.Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
hallucinate
AI systems generating incorrect or fabricated information
intervening layer
A system that sits between the user and the original information source
midterm elections
US elections held between presidential elections, featuring Senate races
Questions readers ask
Which chatbot recommended candidates most often when users shared their views?
ChatGPT recommended candidates in 79% of chats after users shared their views, according to the Yale study.
Did any chatbot recommend candidates without users providing personal information?
Yes, Grok regularly recommended a candidate in 14% of chats where users had not yet said anything about themselves.
How many people participated in the Yale study?
Nearly 2,500 people in 35 states with Senate races participated in the study.
Connected stories
Reading receipt
- One-third of study participants asked chatbots to help select a political candidate.
- Grok recommended candidates in 14% of chats where users provided no personal information.
- ChatGPT issued candidate recommendations in 79% of chats once users shared their views.
Still open: How do these recommendation rates change in non-midterm election cycles?
Still open: What specific source weighting algorithms do these models use when assessing candidate alignment?