In this Letter we revisit arguments suggesting that the Bardeen-Petterson effect can coalign the spins of a central supermassive black hole binary accreting from a circumbinary (or circumnuclear) gas disc. We improve on previous estimates by adding the dependence on system parameters, and noting that the nonlinear nature of warp propagation in a thin viscous disc affects alignment. This reduces the discs ability to communicate the warp, and can severely reduce the effectiveness of disc-assisted spin alignment. We test our predictions with a Monte Carlo realization of random misalignments and accretion rates and we find that the outcome depends strongly on the spin magnitude. We estimate a generous upper limit to the probability of alignment by making assumptions which favour it throughout. Even with these assumptions, about 40% of black holes with $a gtrsim 0.5$ do not have time to align with the disc. If the residual misalignment is not small and it is maintained down to the final coalescence phase this can give a powerful recoil velocity to the merged hole. Highly spinning black holes are thus more likely of being subject to strong recoils, the occurrence of which is currently debated.