Enhancing Democratic Listening in Online Discussions: The Role of Reaction Buttons, Discussion Moderation, and Large Language Models

Political communication research in general – and research on political online discussions in particular – has previously focused on the exercise of democratic voice and thus largely neglected the complementary practice of democratic listening. Against the backdrop of a wide-spread perception that democratic listening is lacking in online discussions, we investigate whether listening can be strengthened through novel platform affordances. We developed an experimental online written discussion platform with listening-oriented reaction buttons and both human and AI moderation (based on GPT-4). We then conducted an experiment involving n = 740 respondents in 40 discussion groups (N = 4,355 comments), each lasting 30 minutes, covering three experimental conditions and a control condition. We compare the effect of (1) listening-oriented reaction buttons versus (2) a combination of these buttons and listening-oriented human discussion moderation versus (3) a combination of buttons and AI moderation on the level of democratic listening demonstrated in discussants’ textual contributions. We find that listening-oriented buttons alone do not enhance democratic listening. However, when combined with listening-oriented AI moderation, they are associated with a democratic listening rate of 27% within discussions, compared to 21% in the condition without either feature. Our findings lay the basis for the implementation of a combination of listening-oriented buttons and discussion moderation on news media websites, social media apps or dedicated discussion spaces using LLM-based chatbots.