This paper explains the need to access chatbot replies based on user preferences to improve interaction and satisfaction. Conventional parameters do not truly reflect the experience of a user, and hence, a user-centric methodology is pivotal. This work introduces a machine learning-based model that predicts preferred responses using features such as relevance, tone, and completeness of the chatbot’s responses. We apply systems trained to predict responses most in accordance with users’ expectations to create more personalized, engaging interactions across the board through random forest models complemented with synthetic minority over-sampling technique (SMOTE) for imbalanced data manipulation. Results obtained show higher user satisfaction and contribute toward conversational AI by bringing it closer to being more adaptive and contextually aware.