This research analyzes language features associated with flexibility and their effect on attitudinal loyalty by using text mining techniques on a dataset of about 5k chat conversations. Two positive constructs (confirmation, owning statements) and two negative constructs (argumentativeness, anxiety) are measured through a top-down method using self-made dictionaries. Confirmative words have a positive effect on attitudinal loyalty, while owning statements have a negative effect on attitudinal loyalty. Moreover, handovers by a chatbot negatively moderates the effect between service providers flexibility and attitudinal loyalty. A post hoc study identifies the most popular topics between service providers and customer, suggesting that the former should use owning statements with caution when talking about topics such as personal information and technical issues