As sentiment analysis and intent recognition technologies have made certain progress, intelligent customer service systems have been widely used. However, existing systems still have difficulty in accurately capturing and efficiently inferring diverse user sentiment expressions and complex intent recognition. To this end, this paper proposes a language strategy optimization framework for intelligent customer service systems based on pragmatic ability analysis, aiming to optimize the language strategy of intelligent customer service systems through in-depth analysis of pragmatic ability. First, the study uses recurrent neural networks (RNNs) to model the input language sequence to capture the temporal dependencies in the context and generate coherent and natural language strategies. In order to enhance the model’s memory and context understanding in long texts, this paper further introduces long short-term memory networks (LSTMs) to effectively solve the problem of information loss in long sequences, thereby improving the accuracy of the generation strategy. At the same time, the Transformer architecture is adopted to optimize the generated language strategy by taking advantage of the efficiency and parallelism of its self-attention mechanism in processing long texts, especially in complex contexts. Experimental results show that language strategy optimization based on pragmatic ability analysis can significantly improve the user experience and service quality of the intelligent customer service system.