Context. The ESA Rosetta mission explored comet 67P/Churyumov-Gerasimenko in 2014−2016, following its target before and after the perihelion passage on 13 August 2015. The NAC camera of the OSIRIS imaging system allowed to map the nucleus surface acquiring images with different filters in the visible wavelength range. Aims. Here we study the spectrophotometric behaviour of the nucleus by a multivariate statistical analysis, aiming to distinguish homogeneous groups and to constrain the bulk composition. Methods. We applied the G-mode clustering algorithm to 16 OSIRIS data cubes acquired on 5−6 August 2014 (mostly covering the northern hemisphere) and 2 May 2015 (mostly covering the southern hemisphere), selected to have complete coverage of the comet’s surface with similar observing conditions. Results. We found four similar homogeneous groups for each of the analysed cubes. The first group corresponds to the average spectrophotometric behaviour of the nucleus. The second (spectrally redder) and the third (spectrally bluer) groups are found in regions that were already found to deviate from the average terrain of the comet by previous studies. A fourth group (characterised by enhancements of the flux at 700−750 nm and 989 nm, possibly due to H2O+ and/or NH2 emissions) seems connected with the cometary activity rather than with the bulk composition. Conclusions. While our aim in this work was to study the spectrophotometric behaviour of the nucleus of 67P/Churyumov-Gerasimenko as a whole, we found that a follow-up application of the G-mode to smaller regions of the surface could be useful in particular to identify and study the temporal evolution of ice patches, as well as to constrain the composition and physical processes behind the emission of dust jets.