Co-crystal design is a convenient way to remedy the poor biopharmaceutical properties of drugs. Most studies focus on experimental co-crystal screening or computational prediction, but hardly any work has been done toward fast, efficient, and reliable prediction of solution crystallization for co-crystal formation. Here, we study the caffeine-benzoic acid co-crystal system, due to its reported difficulty to crystallize from the solution phase. With this work, we investigate whether there is a link between prenucleation aggregation in solution and co-crystal formation and how to harness this for crystallization prediction. 1H and 13C NMR spectroscopy is used to study the prenucleation interaction between caffeine and benzoic acid in methanol, acetone, and acetonitrile as examples of common solvents. In this system, crystallization from methanol leads to no co-crystallization, from acetone to concomitant crystallization of co-crystal and caffeine, and from acetonitrile to pure co-crystal formation from solution. Strong heteromeric dimers were found to exist in all three solvents. Ternary phase diagrams were defined and a solution-accessible co-crystal region was found for all solvents. For this system, the prenucleation clusters found in solution could be linked to the crystallization of the co-crystal. Crystallization from DMSO did not yield the co-crystal and there were no detectable prenucleation aggregates. NMR spectroscopy to probe dimers in solution can thus be used as a fast, reliable, and promising tool to predict co-crystallization from specific solvents and to screen for suitable solvents for manufacturing and scale-up.