The Shape mode calculates all touching parts of one layer and selects them. Since this is achieved independent of the net name, it also works for Gerber and other formats that do not contain net information.
A common use case for the Shape mode are nets that physically are one but have - because of administrative reasons, e.g. because otherwise they would be too large - been divided into two or more nets with their own net names in the schematics. These nets are marked as logically linked through so-called star points which indicate a fully intentional short curcuiting. With the default mode you would only be able to select the divided parts separately while the Shape mode provides you the whole coherent network.