Judged by a Chatbot? An Empirical Investigation of the Impact of Expectancy Violations on Users' Trust in AI-based Chatbots.

Conversational artificial intelligence (AI) is increasingly being incorporated into mental health applications such as chatbot coaches.Since AI-based chatbots are seen as free from human biases or emotional reactions, users tend to expect non-judgemental interactions with them.However, as in interactions with other people, expectations can be violated in interactions with chatbots.Using Expectancy Violations Theory as a framework, this study aims to explore how these violations, in contrast to confirmations, may impact users' trust, feeling of being understood, and connectedness to the chatbot.Due to the specific setting (AI-based chatbot acting as mental health coach), fear of stigma and technical affinity were also considered.In an online experiment with a 2x2 between-subjects design (N = 95), participants interacted with a chatbot that was either described as judgmental or non-judgmental and either displayed a judgmental or non-judgmental communication behavior.Results show that the non-judgmental chatbot behavior results in more positive trust evaluations than the judgmental chatbot behavior.Neither the confirmation nor the violation of expectations was found to have a significant impact.Overall, the chatbot's communication behavior was more decisive for evaluations than users' expectations, technical affinity, or fear of stigma.