<p>This study examines a market that two heterogeneous online retailers sell two substitute products with fluctuating quality to experiential-learning consumers, both without and with AI-assisted consumer decision-making. Without AI, experiential-learning consumers update quality beliefs based on previous purchase experiences, but consumers could learn the true product quality with AI assistance. This study analyzes the dynamic multi-period purchasing behavior of consumers, compares equilibrium about return policy for retailers with and without AI, investigates the impact of AI on retailer revenues, return policy equilibrium, consumer surplus and overall market performance. We make some extensions and analyze the impacts of dynamic pricing and learning from imperfect AI on the return policy equilibrium, and examine the scenario where consumers learn from online reviews. Results indicate that quality fluctuation and the experiential learning leads to a lower consumer quality belief than the actual quality. However, AI effectively corrects consumers’ downward-biased beliefs. For retailers with more volatile fluctuation or higher quality products, AI can boost their revenues, yet may diminish their incentive to offer money-back guarantees, except when the AI has a significantly poor prediction ability. Under certain conditions, AI can significantly improve consumer surplus and increase the total market revenue. Dynamic pricing according to consumer quality belief can also eliminate the negative impact of quality fluctuations. Learning from online reviews helps reduce fluctuations in consumer quality belief.</p>