Using global navigation satellite system (GNSS) data to detect millimeter-order signals of short-term slow slip events (S-SSEs) and to estimate their source parameters, especially duration, is challenging because of low signal-to-noise ratio. Although the duration of S-SSEs in the Nankai subduction zone has been estimated using tiltmeters, its regional variation has never been quantitatively studied. We developed an S-SSE detection method to estimate both the fault model and duration with their errors based on the detection methods developed by previous studies and applied it to a 23-year period of GNSS data in the Nankai subduction zone. We extracted S-SSE signals by calculating correlation coefficients between the GNSS time series and a synthetic template representing the time evolution of an S-SSE and by computing the average of correlation coefficients weighted by the predicted S-SSE signals. We enhanced the signals for duration estimation by stacking GNSS time series weighted by displacements calculated from the estimated fault model. By applying the developed method, we detected 284 S-SSEs from 1997 to 2020 in the Nankai subduction zone from Tokai to Kyushu and discussed their regional characteristics. The results include some newly detected S-SSEs, including events accompanying very low-frequency earthquakes and repeating earthquakes in offshore Kyushu. Our study provides the first geodetic evidence for synchronization of S-SSEs and other seismic phenomena in offshore Kyushu. We estimated the cumulative slip and duration, and their error carefully. We also estimated the average slip rate by dividing the cumulative slip by the cumulative duration. This study clarified that the average slip rate in western Shikoku was approximately twice as that in eastern Shikoku and Kyushu. These regional differences were statistically significant at the 95% confidence interval. Multiple factors can influence the regional characteristics of S-SSEs, and we speculate that the subducting plate interface geometry is one of the dominant factors.