The OSIRIS-REx spacecraft encountered the asteroid (101955) Bennu on December 3, 2018, and has since acquired extensive data from the payload of scientific instruments on board. In 2019, the OSIRIS-REx team selected primary and backup sample collection sites, called Nightingale and Osprey, respectively. On October 20, 2020, OSIRIS-REx successfully collected material from Nightingale. In this work, we apply an unsupervised machine learning classification through the K-Means algorithm to spectrophotometrically characterize the surface of Bennu, and in particular Nightingale and Osprey. We first analyze a global mosaic of Bennu, from which we find four clusters scattered across the surface, reduced to three when we normalize the images at 550 nm. The three spectral clusters are associated with boulders and show significant differences in spectral slope and UV value. We do not see evidence of latitudinal non-uniformity, which suggests that Bennu's surface is well-mixed. In our higher-resolution analysis of the primary and backup sample sites, we find three representative normalized clusters, confirming an inverse correlation between reflectance and spectral slope (the darkest areas being the reddest ones) and between b' normalized reflectance and slope. Nightingale and Osprey are redder than the global surface of Bennu by more than 1σ from average, consistent with previous findings, with Nightingale being the reddest (S' = (-0.3 ± 1.0) × 10-3 percent per thousand angstroms). We see hints of a weak absorption band at 550 nm at the candidate sample sites and globally, which lends support to the proposed presence of magnetite on Bennu.