Recommended, purchased, disappointed?- Exploring the divergent effects of algorithmic product recommendations on sales and satisfaction

With the rise of algorithmic commerce, data-driven product recommendations have become central to online retail. This study investigates whether such recommendations enhance product performance and consumer satisfaction, and how their effects vary by consumption timing (immediate vs. non-immediate) and product characteristics (price and type). Using real-time ‘Gold List’ rankings from JD.com, we analyse over 28,000 product records and consumer reviews. Satisfaction is quantified using a supervised Word2vec-SVM model, and effects are estimated using a double machine learning (DML) framework. Results show that recommendations significantly improve product performance, especially for non-immediate consumption and experience goods, but have limited or even negative effects on satisfaction. Topic modelling (LDA) of negative reviews reveals dissatisfaction drivers such as weak after-sales support and quality concerns. Robustness checks confirm the consistency of findings. This study contributes to understanding when and why algorithmic recommendations succeed or fail, offering insights for personalized retail strategies in AI-driven digital marketplaces.