Computed tomography (CT) images of metal components are often degraded due to artifacts associated with X-ray scattering within dense materials. Such artifacts reduce the image contrast and complicate the detection of tiny defects in the low-contrast regions. To ensure the detectability of such defects using X-ray CT, artificial defects are often embedded in a sample component, which is made equivalently with the real one in its shape and radiological property. Such samples are technically considered a representative quality indicator (RQI). The artificial defects must be embedded at the regions where the contrast in a CT image is low. To determine such low-contrast regions, CT scans have to be repeated, either real or simulated, by changing the positions of artificial defects. The repetition of CT scans is too expensive to justify in practice. More seriously, when using real samples, we need to rework them to correct the positions of artificial defects. Therefore, it is difficult to obtain appropriate positions of the artificial defects. To address this problem, this study introduces an efficient simulation-based approach to recommend the positions of artificial defects by identifying the low-contrast regions. Specifically, we used the contrast-to-noise ratio (CNR) as an indicator to show the detectability of defects and recommend to locate artificial defects at the regions with low CNR values. Our method calculates a CNR value at every voxel using a computer-aided design (CAD) mesh model of the sample without artificial defects. We first compute a CT volume consisting of three components, primary, scatter, and noise and then calculate the contrast value for each voxel. Second, we compute another CT volume that only consisted of noise component. Finally, a CNR value is computed by dividing the contrast by the noise magnitude at each voxel. We tested the proposed method with real metal pieces made by additive manufacturing. By comparing the positions of artificial defects with the ones defined by human engineers in several template models, we found that the proposed system suggested comparable positions.