Context. Molecular gas is a necessary fuel for star formation. The CO (1−0) transition is often used to deduce the total molecular hydrogen but is challenging to detect in low-metallicity galaxies in spite of the star formation taking place. In contrast, the C  II λ 158  μ m is relatively bright, highlighting a potentially important reservoir of H 2 that is not traced by CO (1−0) but is residing in the C + -emitting regions. Aims. Here we aim to explore a method to quantify the total H 2 mass ( M H 2 ) in galaxies and to decipher what parameters control the CO-dark reservoir. Methods. We present Cloudy grids of density, radiation field, and metallicity in terms of observed quantities, such as O  I , C  I , CO (1−0), C  II , L TIR , and the total M H 2 . We provide recipes based on these models to derive total M H 2 mass estimates from observations. We apply the models to the Herschel Dwarf Galaxy Survey, extracting the total M H 2 for each galaxy, and compare this to the H 2 determined from the observed CO (1−0) line. This allows us to quantify the reservoir of H 2 that is CO-dark and traced by the C  II λ 158  μ m. Results. We demonstrate that while the H 2 traced by CO (1−0) can be negligible, the C  II λ 158  μ m can trace the total H 2 . We find 70 to 100% of the total H 2 mass is not traced by CO (1−0) in the dwarf galaxies, but is well-traced by C  II λ 158  μ m. The CO-dark gas mass fraction correlates with the observed L C  II / L CO(1−0) ratio. A conversion factor for C  II λ 158  μ m to total H 2 and a new CO-to-total- M H 2 conversion factor as a function of metallicity are presented. Conclusions. While low-metallicity galaxies may have a feeble molecular reservoir as surmised from CO observations, the presence of an important reservoir of molecular gas that is not detected by CO can exist. We suggest a general recipe to quantify the total mass of H 2 in galaxies, taking into account the CO and C  II observations. Accounting for this CO-dark H 2 gas, we find that the star-forming dwarf galaxies now fall on the Schmidt–Kennicutt relation. Their star-forming efficiency is rather normal because the reservoir from which they form stars is now more massive when introducing the C  II measures of the total H 2 compared to the small amount of H 2 in the CO-emitting region.