We provide a framework for analyzing search across correlated observations. The agent—an online customer, a drug company, a politician—tracks innovations over a Brownian path. The agent chooses the speed and length of search and retrospectively picks the best innovation when search is completed. We show that the optimal search speed is U-shaped: it is highest when approaching a breakthrough or when nearing search termination. Unlike search across independent samples, search optimally stops when observations are sufficiently discouraging, following a drawdown stopping boundary. We also show the tractability and features of optimal search contracts in our setting.