To clarify the roles and application pathways of large language models in intelligent digital services, this study reviews relevant research through the lens of information processing theory and summarizes application patterns across users, service providers, platforms, and multi-actor interactions. The findings reveal a progressive intelligence paradigm comprising perceptual, cognitive, reasoning, and execution intelligence. Major approaches include prompt engineering, fine-tuning, retrieval-augmented generation, and reinforcement learning, while key challenges involve intent understanding, knowledge retrieval, reasoning hallucinations, and multi-agent coordination. These challenges hinder the translation of model capabilities into service effectiveness, requiring coordinated support in information perception, knowledge cognition, logical reasoning, and action execution.