In semiempirical ionic models a number of adjustable parameters have to be fitted to experimental data of either monomer molecules or crystals. This leads to strong correlations between these constants and prevents a unique test and a clear physical interpretation of the fit parameters. Moreover, it is not clear whether these constants remain unchanged when the model is applied to dimers or larger clusters. It is shown that these correlations can be substantially reduced when reliable information about dimers is available from experiments or ab initio calculations. Starting with Dunham coefficients of the monomer potential determined from microwave measurements, we have calculated the monomer to dimer bond expansion and the bond angle without any additional adjustable parameter. Assuming that the overlap repulsion between nearest neighbors remains unchanged, the bond expansion is mainly determined by the simple Coulomb repulsion between equally charged ions and depends only very little on the effective ion polarizabilities. Deviation of the bond angle from 90 degrees sensitively tests the difference of effective polarizabilities of the two ions. A comparison with previously available data and new ab initio MP2 results presented here for the heavy-atom containing dimers shows that bond angles can be modeled reasonably well with Seitz-Ruffa corrected Pauling polarizabilities while calculated bond expansions are much too long, This shows that changes of the overlap repulsion term must be considered for reliable predictions of the structure of dimers and larger clusters.