Surprisingly, they don't, at least as far as I know. I haven't ever heard of anybody giving, or even trying to give, a proper definition of a maxim, in particular of the level at which it is to be stated (that is underspecified, if not to say unspecified, which makes the whole categorical imperative extremely vulnerable to rationalizations), and of the way that the description of the hypothetical situation in which the maxim is universalised is to be computed. My suspicion, though I haven't done any research to confirm it, is that this is because philosophers who like Kantian ethics don't like formal logic and have no clue about causal models and counterfactuals.