Chatbot Voting Advice Applications inform but seldom sway young unaligned voters

Abstract

Voting Advice Applications (VAAs) are interactive tools that communicate information about elections, yet their effectiveness in enhancing political knowledge and participation remains understudied. Moreover, traditional VAAs may disproportionately attract politically engaged users with already well-formed ideological views, limiting their potential to inform a broader and less engaged electorate. This paper introduces a “VAA Bot” that employs large language models (LLMs) and retrieval-augmented generation to deliver balanced, personalized information drawn from official party platforms and public documents. We evaluate the VAA Bot’s impact across three experimental studies aimed at young politically unaffiliated adults. The findings provide evidence that the VAA Bot improves knowledge of party stances on issues of great importance to each user. However, the VAA Bot produces weak effects on downstream outcomes such as vote preferences and party evaluations among respondents whose primary issue position aligns closely with one of the parties. These findings contribute to ongoing debates about the role of political information in shaping behavior and clarify both the promise and the limitations of LLM-based tools for civic learning.