At temperatures and chemical compositions approaching the solidus line in binary alloys, many grain boundaries (GBs) become increasingly disordered and pre-melt by developing a thin liquid layer. The phenomenon can be described by a temperature and composition dependent disjoining potential characterizing the interaction between the two solid-liquid interfaces bounding the liquid layer. We present a theory describing probability distributions and statistical correlations between of the GB width, excess energy and segregation arising as a result of thermal fluctuations. It is demonstrated that critical information can be extracted from the fluctuation analysis, including the disjoining potential The proposed theory is applied to semi-grand-canonical Monte Carlo simulations of GB premelting in Cu-rich alloys of the eutectic Cu–Ag system modeled with an embedded-atom potential. For different grain boundaries, the disjoining potential can be attractive, repulsive, or mixed depending on the alloy composition and temperature.