The integration of generative artificial intelligence into Clinical Decision Support Systems (CDSS) represents a promising direction for improving patient interaction and the quality of input data. This study explores the potential role of ChatGPT in the collection and structuring of patient medical histories through two applications: the direct classification of free-text medical histories, and the use of ChatGPT as an interactive agent to assist patients in providing medical information via natural language dialogue. Experimental results show that the direct classification of unstructured medical histories by ChatGPT yields suboptimal performance, highlighting the limitations of general-purpose language models in specialized clinical prediction tasks. Conversely, the interaction with ChatGPT to guide patients in narrating their medical history, followed by the structuring of this information, produces outputs that can be more effectively integrated into downstream classification modules. These findings suggest that, rather than serving as an autonomous decision-making system, ChatGPT could play a supportive role in facilitating early-stage data acquisition for clinical workflows.