We conducted a year-long experiment among millions of new sellers on an e-commerce platform. The treatment group received access to a free and customized entrepreneur training program, in which an AI algorithm assigned training materials to sellers based on their real-time operating statistics. With a 24% take-up rate, new sellers eligible for training saw 1.7% higher revenue on average (6.6% ATT), primarily driven by higher traffic and enhanced advertising. To understand the welfare impact, we present an equilibrium model in which consumers search sequentially based on platform ranking, while new sellers can advertise at a cost to signal their quality, subject to information friction. Using panel data of consideration sets, we validate model predictions and document how training improved screening among new sellers, expanding the market. Overall, this study reveals the information friction on--and the signaling value of--advertising among platform-dependent entrepreneurs.