We investigated temporal variations of seismic wave velocity associated with hydraulic fracturing using Green's functions computed from ambient noise data. In October and November of 2014, we set up a broadband array at a shale play site inside the Sichuan basin where a pilot horizontal drilling and hydraulic injections were conducted. We first computed cross‐correlation functions using continuous data recorded by 21 three‐component broadband sensors deployed around the treatment well. We then employed a running window correlation‐based coda wave interferometry technique to measure apparent velocity changes from the daily Green's functions of all the station pairs in the frequency range of 1 to 3 Hz. We found significant velocity changes right after the hydraulic fracturing, which exhibited a clear direction‐dependent pattern. S wave velocity along raypaths parallel to the well trajectory showed a clear increase while those perpendicular exhibited a small decrease. The anisotropic changes in seismic velocity observed here were also confirmed from surface wave horizontal particle motion data. By comparing our observations with normal stress changes calculated with a half‐space elastic model, we speculate that stress changes induced by the hydraulic fracturing were likely to be responsible for the observed anisotropic changes in seismic velocity. Our results suggest that time‐lapse seismic imaging with ambient noise data provides a promising probe for monitoring geomechanical changes related to exploitation of unconventional oil and gas resources.