Online stores actually need those cranky customers who leave bad reviews to keep the whole rating system from becoming a joke.
SSRN · March 17, 2026 · 6300378
The Takeaway
If only fans review a product, the descriptions become homogenized and lose their information value; the paper finds a 'diversity dividend' where unhappy consumers provide the necessary details that help others determine product fit. Paradoxically, highly effective AI recommendation algorithms can destroy a platform's value by ensuring only 'matching' customers buy the product, thereby killing the information flow.
From the abstract
Consumer reviews are predominantly studied as quality signals-tools for sorting products from best to worst. But they also serve a routing function, helping consumers assess not whether a product is good but whether it is good for them. We model review text as a signal that different readers decode heterogeneously, generating personalized inferences without personalized delivery. As routing succeeds, the buyer pool homogenizes and the text signal degrades. The unique equilibrium features partial