Early marriage is one of the social issues in Indonesia, with a significant negative impact on child development and community welfare. Despite various efforts to reduce early marriage, its prevalence remains high, particularly in regions with low education levels, high poverty, and strong social norms. Artificial Intelligence (AI) and Natural Language Processing (NLP) technologies have opened opportunities to develop AI assistants. However, existing AI assistants often only provide factual answers without showing empathy or offering safe, convenient, and effective solutions. In this study, we present the application of the Reinforcement Learning (RL) approach to evaluate persuasive sentences for early marriage prevention in AI assistants. The proposed system integrates Proximal Policy Optimization (PPO) and Deep Q-Network (DQN) algorithms to optimize the interaction of AI assistants in providing more effective and empathetic persuasive messages. The results show that the Deep Q-Network (DQN) algorithm provides better performance with a score of 0.962, compared to Proximal Policy Optimization (PPO), which achieves a score of 0.715. We used the Unified Multi-Dimensional Evaluator (UniEval) evaluation matrix to measure the quality of the results with a focus on consistency.