Social bots have emerged as sophisticated agents in online communities, serving as a practical lens through which the societal implications of artificial general intelligence can be explored. These bots can participate in various positive social interactions, including the provision of entertainment services and community moderation, but some bots have negative effects, such as spreading misinformation. Therefore, to maintain the harmony of online communities, it is crucial to distinguish between “good” and “bad” bots and to identify the characteristics that can be used to make this distinction. Using the elaboration likelihood model, this study integrates text and sentiment analysis with machine learning algorithms to explore what makes bots good from a user perspective. Drawing on data from Reddit, we analyze how the textual sentiment and textual paralanguage embedded in bot-generated comments drive positive social interactions. Our results indicate that textual sentiment, processed via the central route, has a stronger impact on user evaluations than textual paralanguage, which operates through the peripheral route. Overall, our study offers important insights into bot design and the factors that promote positive social interactions within online communities, thus supporting platform governance.