Do Voters Punish Candidates Who (Mis-)Use Artificial Intelligence?

This project examines how voters evaluate political candidates who use artificial intelligence (AI) in their election campaigns, and whether they are willing to trade off the misuse of AI against perceptions of candidate competence. We conduct a large-scale cross-national conjoint experiment across ten countries: the United States, United Kingdom, Germany, Italy, Czech Republic, South Korea, Mexico, India, Nigeria, and South Africa. The sample is designed to capture variation in political regimes, democratic trajectories, regional diversity, and levels of AI adoption. In each conjoint task, respondents are presented with profiles of two hypothetical political candidates that vary across eight attributes: age, gender, education, profession, party affiliation, use of AI in the campaign, competence in handling the economy, and competence in fighting corruption. The AI attribute spans a realistic spectrum, from no use and neutral applications (drafting campaign texts, tailoring messages to voter groups) to clearly deceptive misuse (fabricated or compromising audio and video of opponents). Respondents indicate which candidate they would be more likely to support in an election and separately rate each candidate on a 1–10 likelihood-of-support scale. We estimate average marginal component effects (AMCEs) to identify how different forms of AI use affect electoral support, and average component interaction effects (ACIEs) to examine whether voters are ready to forgive AI misuse if a candidate is competent. The study addresses two main questions: whether voters distinguish between neutral and deceptive uses of AI in campaigns, and whether individuals are willing to tolerate or forgive AI misuse in exchange for higher perceived competence in handling the economy or fighting corruption. The findings contribute to debates on the electoral consequences of AI in politics and shed light on the conditions under which voters sanction or tolerate digital deception.