We show that an event-shape engineering based on the mean transverse momentum of charged hadrons, $[p_t]$, provides an optimal handle on the strength of the magnetic field created in central heavy-ion collisions at high energy. This is established through quantitative evaluations of the correlation existing between the event-by-event magnetic field produced by the spectator protons in 5.02 TeV Pb+Pb collisions and the event-by-event $[p_t]$ at a given collision centrality. We argue that the event selection based on $[p_t]$ provides a better handle on the magnetic field than the more traditional selection based on the event ellipticities. Advantages brought by this new method for the experimental search of the chiral magnetic effect are discussed.