Optimum amplification in Surface Enhanced Raman Scattering (SERS) from individual nanoantennas is expected when the excitation is slightly blue-shifted with respect to the Localized Surface Plasmon Resonance (LSPR), so that the LSPR peak falls in the middle between the laser and the Stokes Raman emission. Recent experiments have shown when moving the excitation from the visible to the near-infrared that this rule of thumb is no more valid. The excitation has to be red-shifted with respect to the LSPR peak, up to 80nm, to obtain highest SERS. Such discrepancy is usually attributed to a Near-Field (NF) to Far-Field (FF) spectral shift. Here we critically discuss this hypothesis for the case of gold nanocylinders. By combining multi-wavelength excitation SERS experiments with numerical calculations, we show that the red-shift of the excitation energy does not originate from a spectral shift between the extinction (FF) and the near-field distribution (NF), which is found to be not larger than 10nm. Rather, it can be accounted for by looking at the peculiar spectral dependence of the near-field intensity on the cylinders diameter, characterized by an initial increase, up to 180nm diameter, followed by a decrease and a pronounced skewness.