The effect of chatbot-mistakes on the user satisfaction with a human-like versus non-human-like chatbot

Abstract

Chatbots are becoming more popular as they are helpful in a variety of fields. Their development is continuously growing. In this research we want to determine whether making a chatbot more human-like can result in a better user satisfaction. We also want to test the effect that mistakes have on user satisfaction and if people are more forgiving towards a human-like chatbot that makes mistakes than a non-human-like chatbot that makes mistakes. In order to do so, four chatbot versions were designed: a human-like version, a non-human-like version, a human-like version that makes mistakes and a non-human-like versuin that makes mistakes. For this research we used the TAM with added categories of social presence and emotional design. We used a questionnaire to evaluate the responses. ANOVA and Pearson correlation were used to analyse the gathered data. We found significant results in both the quantitative and qualitative data, and found that a human-like chatbot does improve the user satisfaction. We also found evidence in qualitative data that users are more forgiving towards a human-like chatbot than towards a non-human-like chatbot.