In existing post-tensioned prestressed concrete bridges, the unfilled grout areas in the prestressed concrete ducts have become a critical issue. These areas are impossible to inspect directly through visual observation, necessitating non-destructive evaluation methods. Therefore, a non-destructive evaluation method is sought for the inspection. Post-tensioned PC structures are commonly employed in bridges with long spans, and a two-stage inspection comprising efficient scanning and detailed examination of the PC duct can be practical. In this study, we developed a scanning method to judge the grouting condition of the duct in post-tensioned PC structures using impact testing and AI classifiers. Three types of specimens were prepared: Two base specimens (referred to as Specimens A), two specimens obtained from a dismantled actual bridge (referred to as Specimens B), and a specimen extracted from a different bridge than B (referred to as Specimen C). A signal processing filter utilizing parasitic discrete wavelet transform (P-DWT) was constructed with Specimens A. Training data were accumulated from Specimens B, the data of which were processed by P-DWT, and a machine learning classifier was created. A test was conducted on Specimen C, and the results were discussed for accuracy by comparing them with the results of the radiographic transmission test. As a result, the classifier based on the random forest achieved an accuracy rate of 4/6 in the test evaluation.