Risk registers are fundamental decision-making tools in project management, yet they frequently suffer from vague, incomplete, or poorly structured entries that undermine effective risk response planning. This paper presents a novel framework that leverages Large Language Models (LLMs) to augment human expertise in risk documentation, addressing a critical gap at the intersection of AI and decision science in managing project uncertainty. Unlike existing AI applications in risk management that focus on prediction or classification, our approach targets the semantic quality of risk descriptions—a previously unexplored area that directly impacts decision-making effectiveness. The framework employs structured prompting strategies with GPT-4 to systematically evaluate risk entries across five dimensions: root cause specificity, incident clarity, impact quantification, strategy alignment, and action concreteness. Applied to a controlled dataset of 20 risk entries, the system demonstrated robust capabilities in detecting semantic deficiencies and generating standardized reformulations. Key findings include: (1) 60% of entries contained hypothetical rather than factual root causes, (2) 55% lacked clear temporal triggers for risk events, and (3) 75% exhibited vague or missing actionable responses. The LLM-based system successfully reformulated 100% of flagged entries to include all required components while maintaining semantic coherence. This research contributes to decision science by demonstrating how AI can enhance, rather than replace, human judgment in managing complex project uncertainties. The framework preserves human oversight while automating the detection and correction of common documentation errors, thereby improving the quality of information available for risk-based decision-making. Future work will focus on empirical validation with industry practitioners and integration with existing project management systems.