We propose preprocessing techniques that improve efficiency of testing language emptiness of alternating automata. We target mainly automata that come from practical application such as string solving or processing of LTL formulae. Such automata often have large alphabets and transition relations represented by succinct and complex Boolean formulae that are more or less unrestricted and may even mix symbols with states. Our main contribution are simplification methods that can be seen as originating from a limited form determinisation, and we also use some simple Boolean transformations of the transition formulae. Our transformations simplify the transition formulae and decrease the number of states. We evaluate experimentally that our preprocessing is beneficial when used together with most of the existing algorithms. It generally improves running times and even allows to solve examples that could not be solved before within a reasonable timeout.