It is a reasonably good predictor only for animals which are relatively closely related and only when in none of the two lineages there has been any significant modification of the way of life, like starting to eat a different kind of food or adapting to an environment where different locomotion or sensory skills are needed.

In this case the accumulated genetic distance consists mainly in random mutations with no or little adaptive value, for which the approximation of a constant "molecular clock" is valid.

The attempts to extrapolate this "molecular clock" hypothesis to other circumstances have produced a huge number of ridiculous research papers whose results have low usefulness.

The approximation of a constant "molecular clock" may be also be applicable in other cases then what I have mentioned in the beginning when the genetic distance does not refer to the complete genome, but only to some groups of genes that have not been involved in the specific adaptations which have been caused in one or both lineages by changes in the way of life, so they have still evolved with a constant mutation rate.

The approximation of a constant "molecular clock" fails in many cases because even if mutations really happen at a relatively constant rate, most of them lead to immediate extinction of the mutated cells and the mutations that we see in the genomes that are examined now are only a small fraction of all mutations, those that have been preserved in survivors.

How large is that fraction depends on where in the genome the mutation has happened, i.e. whether it was in a place with large or small influence on the chances of survival and also on whether the mutated animal lived at that time in an environment with great or little adaptive pressure.