Abstract E-commerce systems increasingly predict which products a customer may click or purchase, but a ranked list rarely explains why an item fits a person’s needs or which trade-offs deserve attention. Generative artificial intelligence (GenAI) creates opportunities for conversational product comparison, review summarization, and personalized guidance; however, fluent output can contain unsupported claims. This paper proposes a conceptual decision-support framework integrating customer-behaviour modelling, purchase-propensity prediction, explainable machine learning, and retrieval-grounded generation. The framework combines searches, views, cart activity, purchases, returns, and review interactions with current product attributes, price, availability, delivery, and return-policy information. A predictive model estimates purchase propensity, an explanation layer identifies relevant evidence, and a generative layer converts it into concise answers with source links and uncertainty disclosures. Trust is treated as an outcome to evaluate—not an automatic benefit of AI—and is supported through transparency, privacy safeguards, user correction, and control over personalization. The paper presents an architecture, research questions, an experimental protocol, and metrics for prediction, calibration, explanation faithfulness, factual grounding, and user understanding. No dataset or user study is claimed; this is a research framework, not a report of measured performance. Its goal is to help shoppers make informed decisions rather than simply maximize conversion.