Most main sequence massive stars (~70%) belong to short-period binaries (P<1 month), a fact that does not reflect the binary parameters measured among populations of newly born massive stars, where the periods exceed years. To understand how such close systems are formed, we need to obtain strong constrains on the origin of the pairing mechanism and the birth orbital properties. Different scenarios have been proposed to produce close binaries, such as the migration that can occur in the framework of disk fragmentation, in which massive binaries are originally formed at large separations and then harden on a time-scale of ~2 Myr. Testing this scenario requires to detect a significant number of relatively massive companions at separations corresponding to the expected size of the accretion disk.
I will describe how optical interferometry (VLTI/GRAVITY) and high-angular resolution techniques (VLT/NACO) are of great importance in characterizing multiplicity at birth. I will show the results using two different sets of observations. I will present the multiplicity and companion fraction for each dataset and compare these results with other studies. I will also discuss the connection with the current star formation theories and show some perspectives of work.