This study evaluates the effectiveness of AI prompt engineering in improving access to safety management information in upstream oil and gas operations. An internal AI system was configured to retrieve company HSE documentation, and its performance was assessed using 30 safety-related prompts across six operational domains, scored on relevance, clarity, and completeness. Results showed a trimodal distribution: 60% perfect scores, 30% complete failures due to indexing/contextual issues, and 10% moderate performance. A key finding was that high clarity scores sometimes accompanied incorrect citations, meaning the AI could convincingly present wrong information. Directly relevant to PSM because it demonstrates both the promise and risk of using prompt-driven AI to access critical safety knowledge in hazardous operations, and highlights the need for rigorous prompt design and output validation.
