Abstract
As generative artificial intelligence becomes increasingly accessible, its potential to reshape political communication poses urgent questions for democratic governance. This thesis examines how exposure to AI-generated campaign materials affects voter behavior, with particular attention to vote choice, emotional response, and perceived authenticity. Using a conjoint experiment administered to 868 U.S. adults via CloudResearch, participants evaluated pairs of campaign flyers — half generated by ChatGPT, half designed by humans — in a simulated state legislative primary. Respondents were not informed that AI-generated materials might appear, replicating real-world conditions in which the origins of political content are frequently obscured.
Results indicate that AI-generated flyers increased the probability of candidate selection by 6.7 percentage points relative to human-authored flyers (p < 0.001, 95% CI: [3.9, 9.6]) — an effect size exceeding those associated with candidate race, gender, and issue salience in the same sample. Contrary to expectations, this advantage operated not through enhanced perceived authenticity, which showed no significant difference between conditions, but through emotional mechanisms: AI flyers generated significantly higher hopefulness and lower worry among respondents. Subgroup analysis reveals a striking interaction with political interest — respondents reporting high political engagement showed an AI effect of 10.5 percentage points, compared to a non-significant 1.5 percentage points among less interested respondents, challenging the conventional assumption that politically engaged citizens are most resistant to persuasive manipulation.
These findings suggest that AI's influence on electoral behavior does not depend on deception. Even respondents who suspected AI involvement remained susceptible to its persuasive effects, raising doubts about the sufficiency of mandatory disclosure laws as a regulatory remedy. This study contributes the first conjoint experimental evidence of AI authorship effects on vote choice and calls for renewed attention to the emotional mechanisms through which AI-generated political content shapes democratic participation.