In an increasingly digital landscape, companies rely heavily on maintaining consistent branding and communication standards, yet the manual effort required to adhere strictly to complex style rules is substantial and errorprone. While rule-based systems effectively handle straightforward cases, their rigidity often fails in nuanced or context-sensitive scenarios. Conversely, LLMs provide remarkable flexibility but struggle to maintain precise consistency when enforcing detailed guidelines.

Through a case study at Rotstift AG, this research first identifies the most time-consuming aspects of style guideline application, revealing that rules involving product naming conventions, hyphenation, and typographic precision constitute significant workflow bottlenecks. A hybrid prototype combining regex-based logic and targeted LLM prompts was developed to address these identified pain points. The prototype underwent iterative testing and refinement, demonstrating substantial improvements in both precision and recall over purely LLM-driven approaches.

A comprehensive evaluation involving real corporate documents and carefully seeded test scenarios demonstrated that the prototype can effectively assist proofreaders by highlighting potential style guideline violations with high accuracy and minimal false positives. The thesis contributes practical insights and methodologies for businesses seeking efficient, flexible, and robust solutions for corporate style guideline automation.