Humans with cognitive ability are able to identify pertinent characteristics in other people’s discussions easily and respond to conversations containing characteristics that already belong to a certain person. However, utilizing personality information for personalized response generation remains a non-trivial task. The system must consider both the user’s conversation history and personality description, posing challenges for coherent model training. In this work, we meticulously design a feature-prompted learning-based dialog generator. This innovative tool primarily harnesses personality description data and dialog history to generate responses. FPLGen uses a clustering mechanism to organize personality description texts into distinct, sparsely populated categories. These categories are merged with historical contextual information and transformed via conditional variational autoencoders. The system incorporates our unique information enhancement and feature-prompted learning strategies, enabling comprehensive dialog synthesis. To validate our model’s efficacy, we conducted experiments on the Chinese persona chat dataset. The results, compared to baseline models, provide irrefutable evidence that our FPLGen model excels at producing richer, more engaging personalized responses.