We propose a simple modification to the recently proposed compressive binary search. The modification removes an unnecessary and suboptimal factor of log log n from the SNR requirement, making the procedure optimal (up to a small constant). Simulations show that the new procedure performs significantly better in practice as well. We also contrast this problem with the more well known problem of noisy binary search.