With increasing product complexity and shorter product lifecycles, manufacturing companies—especially small and medium-sized enterprises (SMEs)—need to adapt their production planning processes quickly and reliably. A key step is the creation of accurate and easily interpretable work instructions for manual assembly tasks. Using Controlled Natural Language (CNL) for these instructions offers distinct advantages: it preserves the readability of natural language while enforcing a standardized syntax and vocabulary. In this context, recent advances in Large Language Models (LLMs) provide an opportunity to streamline the generation of CNL-based assembly instructions. This paper presents a proof of concept that investigates whether an LLM, coupled with a Retrieval-Augmented Generation (RAG) approach, can generate CNL-compliant instructions from Product-Process-Resource (PPR) models. Two representative use cases demonstrate the potential of this approach to improve structural consistency and reduce manual effort. However, the evaluation remains limited to a small number of examples and a single LLM, and current PPR models lack certain critical information such as precise positioning. These findings indicate that while LLMs show promise for supporting automated generation of CNL-based instructions, further work is needed to refine domain adaptation, improve data quality, and validate the approach with broader datasets and alternative language models. This work thus lays the groundwork for more flexible, automated production planning processes that exploit the synergy of LLMs and controlled linguistic frameworks.