When More Experience Is Not More Helpful: Evidence from Positive and Negative Reviews on Steam

As user reviews grow on online platforms, separating useful feedback from noise becomes operationally crucial for guiding consumers and for triaging feedback internally (e.g., for product learning and quality assurance). Yet the most direct indicator of review helpfulness—readers’ votes—arrives later, so platforms must assess the expected helpfulness of new reviews based on cues available at submission. Many platforms display reviewer experience cues, but it is unclear how such experience translates into perceived review helpfulness and whether the relationship differs for positive versus negative reviews. Using 26.8 million video-game reviews on Steam, where reviews display reviewers’ cumulative playtime and a binary verdict (recommend/not recommend), we aim to address this question by estimating models with rich controls and fixed effects. We find that experience predicts helpfulness in a non-monotonic way, with opposite curvature by review verdict: U-shaped for recommended reviews (i.e., reviews from low- and high-experience users receive more helpful votes than reviews from moderately experienced users) but inverted-U for not-recommended reviews (peaking at moderate experience). Observable cues (e.g., review length, product maturity) systematically moderate these patterns. Our findings offer clear guidance for review management: rather than treating experience as a uniform quality signal, platforms should interpret and weigh experience differently based on review verdict and observable context cues.