The resonance conditions of localized surface plasmon resonances (LSPRs) can be perturbed in any number ways making plasmon nanoresonators viable tools in detection of e.g. phase changes, pH, gasses, and single molecules. Precise measurement via LSPR of molecular concentrations hinge on the ability to confidently count the number of molecules attached to a metal resonator and ideally to track binding and unbinding events in real-time. These two requirements make it necessary to rigorously quantify relations between the number of bound molecules and response of plasmonic sensors. This endeavor is hindered on the one hand by a spatially varying response of a given plasmonic nanosensor. On the other hand movement of molecules is determined by stochastic effects (Brownian motion) as well as deterministic flow, if present, in microfluidic channels. The combination of molecular dynamics and the electromagnetic response of the LSPR yield an uncertainty which is little understood and whose effect is often disregarded in quantitative sensing experiments. Using a combination of electromagnetic finite-difference time-domain (FDTD) calculations of the plasmon resonance peak shift of various metal nanosensors (disk, cone, rod, dimer) and stochastic diffusion-reaction simulations of biomolecular interactions on a sensor surface we clarify the interplay between position dependent binding probability and inhomogeneous sensitivity distribution. We show, how the statistical characteristics of the total signal upon molecular binding are determined. The proposed methodology is, in general, applicable to any sensor and any transduction mechanism, although the specifics of implementation will vary depending on circumstances. In this work we focus on elucidating how the interplay between electromagnetic and stochastic effects impacts the feasibility of employing particular shapes of plasmonic sensors for real-time monitoring of individual binding reactions or sensing low concentrations – which characteristics make a given sensor optimal for a given task. We also address the issue of how particular illumination conditions affect the level of uncertainty of the measured signal upon molecular binding.