Abstract:  Voters frequently struggle to understand political attitude statements in voting advice applications (VAAs) and often invest limited effort in resolving these difficulties. Conversational agent VAAs (CAVAAs) aim to reduce the cognitive effort involved in searching for relevant information by integrating chatbots that can provide contextual support. This article presents findings from three studies comparing CAVAAs to standard VAAs without additional information (Studies 1 and 2) and to VAAs with static clickable information (VAA+, Study 3). Study 1 (N = 93) was a laboratory experiment conducted during the 2023 Dutch parliamentary elections. University students were assigned to a standard VAA or a CAVAA. The chatbot in the CAVAA was used in approximately 45% of cases, with users showing a preference for opinion-based and status quo information. Compared to the VAA, CAVAAs reduced non-directional responses to the VAA statements and increased evaluations of the tool’s usability. Moreover, users’ perceived knowledge was higher, whereas no differences were found for factual knowledge and turnout intention. Study 2 (N = 144) largely replicated these results in a field setting with a more diverse sample and showed that the effects hold across different levels of political sophistication. Study 3 (N = 159), conducted during the 2024 European elections, compared a CAVAA to a VAA+. While VAA+ users requested information more frequently, both tools received similar evaluations, and this finding was again consistent across groups of different political sophistication. In the manuscript, we discuss the implications of these findings for theory and practice.

Keywords:  conversational agents; political knowledge; political sophistication; voting advice applications

Published:   11 February 2026

Issue:   Vol 14 (2026): Voting Advice Applications: Methodological Innovations, Behavioural Effects, and Research Perspectives (In Progress)