We propose a new simple technique to improve the short term stability and the robustness of Common-Views (CV) of Global Positioning System (GPS) satellites. We use the ionosphere free linear combination of P-code data broadcast on both GPS carriers, called TAIP3, the data files being built according to the Common GNSS Time Transfer Standard (CGTTS) format. Instead of averaging the outliers out by using a given filter around a mean value computed for each CGTTS sampling epoch, we average the outliers out from a linear fit on daily CV files. We analyze the results over more than 200 d of data. Over a short baseline between remote stations, we see that the average number of satellites left by the daily filtering for the CV computation is almost twice as large as the number obtained from the epoch filtering, increasing that way the robustness of the time transfer. On the short term stability of the time transfer, the white phase noise modulation is improved by a factor larger than the square root of 2, above what was expected. Over a long baseline, the improved short term stability is about twice smaller between 10 000 s and 1 d, together with an average increase of more than 20 per cent on the number of satellites left after filtering for the CV computation. We study the singular event which causes the discrepancy between these two criteria, and conclude that the daily approach is providing the best results in all cases. This automated filtering technique might be useful for any GNSS time transfer based on code data.