Generative Adversarial Networks (GANs) consist of two neural networks a generator and a discriminator that compete in a minimax game: the generator attempts to create realistic text, while the discriminator aims to distinguish between real and generated content. Although effective in many generative tasks, traditional GANs face challenges in text generation, such as maintaining coherence, contextual relevance, and sentiment alignment. To address these limitations, this paper presents a Reinforcement Learning-enhanced GAN (RL-GAN) framework for controlled text generation. The proposed model incorporates sentiment analysis and evaluation metrics BLEU, ROUGE, and METEOR into the reward function to guide the generator toward producing fluent and sentiment-consistent outputs. The model is trained on a large-scale news dataset, where the generator is initially pretrained using Maximum Likelihood Estimation (MLE) and subsequently fine-tuned using policy gradients. A multi-aspect discriminator evaluates generated text based on sentiment, fluency, and topical relevance, providing more nuanced feedback. Experimental results show that RL-GAN outperforms the baseline GAN, achieving higher BLEU-4 (24.2%), ROUGE-1 (37.5%), ROUGE-2 (19.3%), and METEOR (21.6%) scores. Training curves indicate better convergence and reduced overfitting, while qualitative analysis confirms improved coherence and contextual relevance in generated text.