Associative submanifolds are certain calibrated submanifolds in G2-manifolds. There is the hope that counting them will reveal subtle information about the underlying G2-structure. On the other hand, certain singular associatives can be resolved in exactly three different ways, so a naive count will be meaningless. In this talk, we will define homotopy associatives as cobordism classes of threedimensional submanifolds that are adapted to the G2-structure in a rather weak sense. We will see that a given cobordism class can be interpreted as a homotopy associative in exactly three different ways. This might help us to define a consistent counting scheme even when the naive number of associatives in a given cobordism class changes due to singularities.