Agents in two-sided matching markets use a mix of platform-guided recommendation and decentralized communication to learn about potential matches before matching. This paper studies whether, and how, market designers can use recommendation systems to incentivize effective communication between potential matches in the context of an online freelance labor market for tasks. With data from a major international online labor market, we build and estimate a novel model of freelancing supply and demand, in which firms post single-task online jobs they wish to outsource, freelancers submit monetary bids and cover letters for those jobs, the platform algorithmically ranks the bids on each job, and firms, seeing these recommendations, choose which freelancer to hire. In our model, freelancers pay up-front effort costs when writing cover letters, and the more writing effort they expend, the more informative the resulting signal is. However, when freelancers face congestion, they lower their equilibrium writing effort, hampering firms’ abilities to find the worker with the highest match quality. We use our estimated model to investigate how the platform could redesign the recommender system to lower congestion and thus incentivize more communication effort and improve the quality of realized matches. Specifically we study the equilibrium effects of a recommender system that promotes bids that exhibit more effort.