Aims. Chemical networks used for models of interstellar clouds contain many reactions, some of them with poorly determined rate coefficients and/or products. In this work, we report a method for improving the predictions of molecular abundances using sensitivity methods and ab initio calculations. Methods: Based on the chemical network osu.2003, we used two different sensitivity methods to determine the most important reactions as a function of time for models of dense cold clouds. Of these reactions, we concentrated on those between C and C{3} and between C and C{5}, both for their effect on specific important species such as CO and for their general effect on large numbers of species. We then used ab initio and kinetic methods to determine an improved rate coefficient for the former reaction and a new set of products, plus a slightly changed rate coefficient for the latter. Results: Putting our new results in a pseudo-time-dependent model of cold dense clouds, we found that the abundances of many species are altered at early times, based on large changes in the abundances of CO and atomic C. We compared the effect of these new rate coefficients/products on the comparison with observed abundances and found that they shift the best agreement from 3 × 104 yr to (1-3)× 105 yr.Read less <