Abstract
This article examines emerging practices in large language model (LLM) integration within game writing, focusing on how these technologies reshape narrative design, creative workflows, and professional roles. Drawing on evolving industry experimentation and academic research, it outlines the relationship between traditional game writing and LLM-driven approaches, surveys new forms of interactive storytelling such as conversational NPCs, multi-agent simulations, adaptive commentators, and LLM-based text adventures, and identifies their narrative affordances and constraints. The article analyses core challenges, including hallucination, bias, narrative incoherence, and control, and discusses current strategies to address them, such as fine-tuning, prompt engineering, and new authoring tools that position writers as system-level narrative architects. It argues that LLM integration represents not automation but a reconfiguration of co-creative authorship between writers, machines, and players, and calls for further research into ethical design, player reception, and the evolving responsibilities of narrative professionals in LLM-augmented game development.