Chatbots as Voting Aid Applications

• In Event: Reimagining Democracy: Politics and New Technologies

Abstract

Voting aid applications (VAAs) are a popular way to convey basic information about party platforms and candidate positions; however, little research has rigorously examined their effectiveness as a source of information. Moreover, traditional VAAs may appeal to voters who already possess ideologically coherent views, undermining their capacity to broaden political understanding across the electorate. We design a “VAA bot” that leverages large language models (LLMs) and retrieval augmented generation to provide balanced information about parties drawn from official national and state platforms. Focusing on young independents, we find preliminary evidence that a “VAA bot” improves participants’ identification of where parties stand on issues of interest and increases their affinity towards issue-aligned parties. These findings speak to broader debates in the literature about the role of information in political behavior and highlight the promise of LLMs as civic education tools.

Authors

• Yamil Velez, Columbia University

• Donald P. Green, Columbia University

• Semra Sevi, University of Toronto