The interaction area and defect sites play important roles in determining the gas sensing properties of graphene. In this work, steam etching is exploited to fragment large graphene sheets on 3D graphene hydrogel (GH) to small pieces with increased defect sites. The obtained steamed graphene hydrogel (S-GH) displays three and ten times higher responses to NO2 and NH3, respectively, compared with its unmodified GH counterpart before steaming treatment. Furthermore, the S-GH-based NO2 sensor exhibits a low limit of detection (LOD) of 57 ppb, which is much lower than that of a GH-based NO2 sensor (178 ppb). This work demonstrates a facile hydrothermal steaming method to enhance the performance of graphene-based gas sensor by combining 3D structural design and defect modulation of graphene materials.