Context. Cassiopeia A is one of the best-studied supernova remnants. Its bright radio and X-ray emission is due to shocked ejecta. Cas A is rather unique in that the unshocked ejecta can also be studied: through emission in the infrared, the radio-active decay of Ti-44, and the low-frequency free-free absorption caused by cold ionised gas, which is the topic of this paper. Aims. Free-free absorption processes are affected by the mass, geometry, temperature, and ionisation conditions in the absorbing gas. Observations at the lowest radio frequencies can constrain a combination of these properties. Methods. We used Low Frequency Array (LOFAR) Low Band Antenna observations at 30-77 MHz and Very Large Array (VLA) L-band observations at 1-2 GHz to fit for internal absorption as parametrised by the emission measure. We simultaneously fit multiple UV-matched images with a common resolution of 17 '' (this corresponds to 0.25 pc for a source at the distance of Cas A). The ample frequency coverage allows us separate the relative contributions from the absorbing gas, the unabsorbed front of the shell, and the absorbed back of the shell to the emission spectrum. We explored the effects that a temperature lower than the similar to 100-500 K proposed from infrared observations and a high degree of clumping can have on the derived physical properties of the unshocked material, such as its mass and density. We also compiled integrated radio flux density measurements, fit for the absorption processes that occur in the radio band, and considered their effect on the secular decline of the source. Results. We find a mass in the unshocked ejecta of M = 2.95 +/- 0.48 M-circle dot for an assumed gas temperature of T = 100 K. This estimate is reduced for colder gas temperatures and, most significantly, if the ejecta are clumped. We measure the reverse shock to have a radius of 114 '' +/- 6 '' and be centred at 23:23:26, +58:48:54 (J2000). We also find that a decrease in the amount of mass in the unshocked ejecta (as more and more material meets the reverse shock and heats up) cannot account for the observed low-frequency behaviour of the secular decline rate. Conclusions. To reconcile our low-frequency absorption measurements with models that reproduce much of the observed behaviour in Cas A and predict little mass in the unshocked ejecta, the ejecta need to be very clumped or the temperature in the cold gas needs to be low (similar to 10 K). Both of these options are plausible and can together contribute to the high absorption value that we find.