What if conversational agents’ answers appear correct but are judgmental? Piloting the potential effect on people’s experience

Chatbots are becoming increasingly essential to information retrieval and decision support. Yet, it remains unclear how and if the (perceivable) unfairness of chatbot responses affects user experience (UX) with such tools. A pre-experimental phase involved 10 experts and 30 participants in testing a set of fair and unfair chatbot responses related to a fictional Master’s program. Six pairs of highly discriminable fair and unfair answers to a set of six questions about the Master’s program were included for the experiment. The experimental phase involved 75 participants who interacted with chatbots featuring different appearances (male, female, neutral) and programmed to provide responses to the questions about the Master’s program with varying levels of fairness, namely: completely unfair, completely fair, or partially fair (i.e., 50% fair and unfair answers). UX was assessed before and after the interaction, focusing on satisfaction, trust, competence, and helpfulness. The intention to promote was measured with the Net Promoter Score. Fairness strongly impacted experience, F(2, 72) = 14.47; p <.001, which significantly affected intention to promote, F(3, 71) = 49.2; p <.001. Chatbot fairness did not directly affect the willingness to promote the usage. Appearance and prior experience did not significantly influence UX or usage intention. Fairness plays a crucial role in shaping UX. Unfair elements in responses of chatbots can negatively impact users’ decision to promote the usage, emphasizing the importance of chatbots’ unbiased communication. Nevertheless, people show a certain level of acceptability of judgmental statements if there are no direct consequences of unfairness.