As soon as their extragalactic origins were established, the hope to make Gamma - Ray Bursts (GRBs) standardizeable candles to probe the very high - z universe has opened the search for scaling relations between redshift independent observable quantities and distance dependent ones. Although some remarkable success has been achieved, the empirical correlations thus found are still affected by a significant intrinsic scatter which downgrades the precision in the inferred GRBs Hubble diagram. We investigate here whether this scatter may come from fitting together objects belonging to intrinsically different classes. To this end, we rely on a cladistics analysis to partition GRBs in homogenous families according to their rest frame properties. Although the poor statistics prevent us from drawing a definitive answer, we find that both the intrinsic scatter and the coefficients of the $E_{peak}$\,-\,$E_{iso}$ and $E_{peak}$\,-\,$L$ correlations significantly change depending on which subsample is fitted. It turns out that the fit to the full sample leads to a scaling relation which approximately follows the diagonal of the region delimited by the fits to each homogenous class. We therefore argue that a preliminary identification of the class a GRB belongs to is necessary in order to select the right scaling relation to be used in order to not bias the distance determination and hence the Hubble diagram.