Voters viewed campaign AI more favorably when tied to their own party, study finds

The campaign letter stayed the same. Voters’ judgments changed depending on whose party supposedly used AI to help write it. In an October 30 research release, the University at Albany highlighted an experiment with 3,055 U.S. participants that found more favorable assessments when AI use was attributed to their preferred party rather than the opposition.

The study is by Seulki Lee-Geiller, a researcher at the university’s Center for Technology in Government. Published September 24, 2026, in the Journal of Elections, Public Opinion and Parties, it examines how information about AI use and political affiliation affects perceptions of election integrity.

One letter, four descriptions

Lee-Geiller conducted the online survey experiment in June and July 2024, ahead of that year’s presidential election. Participants were randomly assigned to four groups and shown a hypothetical letter from a presidential candidate. Each group received a different description of how it had been created:

• No AI mention: The letter carried no information saying AI had been involved.
• AI assistance: AI had helped draft the letter, without a party-specific attribution.
• Preferred-party attribution: The participant’s preferred political party had used AI.
• Opposing-party attribution: The opposing political party had used AI.

Participants then evaluated election integrity across four dimensions: fairness, authenticity, freedom from fraud and representation of voters’ interests.

AI disclosure versus a partisan cue

Simply telling participants that AI had helped create the message did not significantly change their views of fairness or authenticity, Lee-Geiller said. It did increase expectations that messages were being targeted to specific audiences.

The study abstract describes a different pattern once party identity entered the picture. Attribution to a respondent’s own party was associated with higher assessments of fairness, authenticity, fraud prevention and representativeness. Attribution to the opposing party brought lower evaluations, particularly on fairness and authenticity.

A governance lesson, with a real-world limit

Lee-Geiller argues that transparency, consistent standards and digital literacy can help people make better-informed judgments about AI-assisted messages, regardless of which party uses them. Those are her recommendations, rather than interventions tested in this experiment.

The exploratory study measured short-term reactions to hypothetical campaign materials, not actual election conditions. It does not establish changes in voting behavior or election outcomes. Lee-Geiller identifies how these perceptions develop through exposure to AI-assisted messaging in real-world campaigns as a subject for further research.