Examining the Role of Generative AI in Online Review Dynamics

ABSTRACT Generative AI has sharply reduced the cost of producing persuasive,human-like text, raising concerns that content on digital platforms may becomeincreasingly vulnerable to strategic manipulation. This paper leveragesdiscrete, plausibly exogenous reductions in AI API prices and the launch of more efficient models as supply-side shocks to the cost of generating reviewcontent. It examines the short-term effects of these shocks on online reviewactivity, a form of user-generated content that has been shown to shapeeconomically consequential outcomes. The aim is to assess whether AI-generatedreviews remain observable despite the substantial filtering efforts undertakenby major online platforms, and to examine the strategic contexts in whichsuch reviews are deployed. Using a difference-in-differences design withroughly 13.8 million business reviews during2023-2024 inthe US market, we compare short-term changes in unverified reviews, which aremore susceptible to manipulation, with verified reviews, which are tied toconfirmed consumer experiences. We find that following these shocks, unverifiedreviews become more negative relative to verified reviews: average ratingsdecline, the share of one-star reviews increases, and the share of five-starreviews decreases. We also find evidence of increases in same-day review-volumespikes, consistent with coordinated or automated activity, and show that mostof the effect is driven by the launch of newer, lower-cost models rather than by API price reductions alone. These resultssuggest that: (1) AI-generated review activity is observable at meaningfulscale despite platform filtering, (2) increasesin AI efficiency appear to shift observable unverified review activity towardmore negative content, consistent with competitive attacks but also potentiallyreflecting AI-assisted consumer complaints, and(3) technological capability and efficiency may matter more than nominal pricereductions, possibly because manipulation occurs within an arms race betweenAI-enabled actors and platform defenses. Overall, the findings provide empirical evidence that AIcapabilities can alter online information environments and pose new challengesfor platform trust, consumer welfare, and business and market dynamics.