Algorithmic Electability: Generative AI, Gendered Political Bias, and LGBTQ Candidates

This study examines whether generative AI systems evaluate political candidates differently based on LGBTQ identity cues, gender presentation, and leadership framing. Using 20 candidate profiles evaluated repeatedly across six large language model platforms, it compares numerical ratings of competence, electability, and political strength. Across observations classified by LGBTQ status, non-LGBTQ candidates received higher average electability ratings (7.26 versus 7.01) and political strength ratings (7.43 versus 7.25), while competence ratings were nearly identical. Evaluations also varied across platforms, suggesting that these patterns are not uniform across AI systems. Profiles featuring conventionally masculine leadership descriptions received higher electability ratings than profiles emphasizing nontraditional leadership, although those comparisons also varied in gender and biographical details. The study introduces “algorithmic electability” to describe differences in AI-generated judgments of candidates’ political viability and argues that emerging digital information intermediaries warrant closer examination within scholarship on political representation and candidate evaluation in contemporary American electoral politics.