This study examines whether LLM-enabled text refinement enhances the persuasiveness of marketing content. We compare three advertising conditions，Baseline Copy, Critique-and-Revision, and Multi-agent Debate，and investigate two underlying mechanisms: perceived argument completeness and emotional arousal. Regression and bootstrap mediation analyses show that both AI-assisted workflows significantly improve persuasive intention relative to the baseline. Critique-and-Revision operates through both cognitive and affective pathways, whereas Multi-agent Debate relies more strongly on argument completeness. These findings provide new insights into how collaborative LLM workflows enhance persuasive communication.