Magnetic nanoparticles have proven to be extremely useful in a broad range of biomedical applications. To ensure optimal efficiency, a precise characterization of these particles is required. Thermal Noise Magnetrometry (TNM) is a recently developed characterization technique that has already been validated against other techniques. TNM offers a unique advantage in that no external excitation of the system is required to drive the measurement. However, the extremely small stochastic signal in the fT range currently limits the accessibility of the method, and a better understanding of the influences of the sample characteristics on the TNM signal is necessary. In this study, we present a theoretical framework to model the magnetic noise power properties of particle ensembles and their signal as measured via TNM. Both intrinsic sample properties (such as the number of particles or their volume) and the geometrical properties of the sample in the setup have been investigated numerically and validated with experiments. It is shown that the noise power depends linearly on the particle concentration, quadratically on the individual particle size, and linearly on the particle size for a constant total amount of magnetic material in the sample. Furthermore, an optimized sample shape is calculated for the given experimental geometry and subsequently 3d printed, giving rise to an 3.5 fold increase in TNM signal from approximately 0.007 to 0.026 pT 2 with less than half of the magnetic material.