Social media plays a critical role in shaping consumer behavior and business performance, yet identifying which content features lead to sustained engagement and organizational outcomes remains a challenge. This study develops an AI-driven framework integrating the Elaboration Likelihood Model (ELM) and Social Influence Theory (SIT) to analyze how multimodal content (text, images, videos) influences user engagement and brand impact. Using a dataset of 5,000 Reddit posts from verified brand-affiliated subreddits, the study applies Natural Language Processing (NLP), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs) for content analysis. Panel Vector Autoregression (PVAR) and Granger causality tests are used to explore bidirectional relationships between engagement behaviors and business indicators such as brand search volume and customer retention. This ongoing research contributes to AI-enhanced social media analytics by offering a data-driven, theory-grounded framework for optimizing content strategies and linking engagement metrics to tangible business outcomes.