Attention and saliency on the internet: Evidence from an online recommendation system

Using high-frequency product-level data from an online retail store, we examine whether consumer choices on the internet are consistent with models of limited attention. We test whether consumers are more likely to buy products that receive a saliency shock when they are recommended by new products. We find a sharp and robust 6% increase in the aggregate sales of existing products after they are recommended by a new product. We also establish that the spillover effects of saliency on products further away in the recommendation network are small, suggesting that returns to search via the recommendation network diminish swiftly. Using a structural model we find that saliency has large effects on consumer consideration but a smaller effect on their subsequent choice, conditional on consideration. Counterfactuals suggest that limited attention disproportionately harms top-selling products but recommendation systems can alleviate these search frictions