Abstract
This study evaluated how large language models (LLMs) generate responses to transportation safety queries when prompted from different user perspectives. Prompt design significantly influences AI (artificial intelligence) outputs, so this research investigated how different prompts affected response content and credibility. A structured framework was used to develop ten transportation engineering questions addressing safety, behavior, and technology topics. Each question was paired with three perspective-specific prompts, resulting in 300 responses collected from ten widely-used AI platforms. Qualitative and quantitative analyses were conducted to assess reference quality, validity, and repetition patterns. A total of 2102 references were generated, of which 962 were verified as credible based on accuracy of authorship, publication date, relevance, and link functionality criteria. The results showed variability across platforms and highlight the importance of prompt structure and user perspective in shaping AI-generated content and using LLMs in domain-specific applications such as transportation safety.