Algorithmically Generated Content and Evaluation Costs: Evidence from Dataset Adoption on Kaggle

Abstract

Algorithmically generated content (AGC)—content created automatically by algorithms to showcase digital products—is increasingly adopted by platforms to reduce the evaluation costs associated with digital goods. Examples include Kaggle’s Kerneler, Coursera Summarizing Tool, and Blinkist Pro, which generate content such as summaries for datasets, online courses, or books. Despite the growing prevalence, the impact of AGC on consumer adoption remains underexplored. This study investigates AGC’s role in adoption, focusing on two key phases: the cold start phase, where user-generated content (UGC) is scarce, and the mature phase, when user votes and reviews accumulate. Conducting econometric analyses on dataset adoption patterns in Kaggle, we first find that AGC generated by Kaggle’s Kerneler helps overcome cold start challenges and boosts early adoption, particularly for datasets with high evaluation costs, such as those lacking detailed descriptions or exhibiting greater complexity. Furthermore, as UGC accumulates (measured through user votes), AGC’s marginal influence on adoption diminishes in the mature phase, as users rely on available UGC for dataset evaluation. These findings highlight AGC’s primary effect in reducing adoption barriers when evaluation costs are high and its declining marginal impact when evaluation costs decline, especially with the availability of UGC. This study contributes to research on evaluation costs and uncertainty mitigators, offering insights into how AGC can facilitate early-stage adoption, especially in the absence of UGC. Practically, it underscores AGC’s potential to address the cold start problem on digital platforms, lowering entry barriers for new products and fostering a more inclusive marketplace while maintaining competition.