Creating ethical conversational agents for voting advice applications: Reflections on making informed choices for design and architecture

This paper examines the use of AI in voting advice applications, specifically by means of enhancing the application with a chatbot as a so-called conversational agent voting advice application (CAVAA). We define five principles that are relevant to the usage of AI in this specific context: independence, correctness, transparency, accessibility, and controllability. We explore why these five principles are especially relevant when AI is used to inform voters on current political issues. Then we demonstrate how different AI-methods score along the axis of controllability, which shows the trade-off between control of the sender versus control of the receiver (or user) of the information. Next to controllability, we also discuss how these methods line up with the other four principles and show that these principles connect to each other, as, for example, a high sender control also relates to high correctness. Lastly, we show how one can take these considerations into account when developing a real-world CAVAA by discussing a case study of the 2026 municipality elections in The Netherlands.