The principle of a new type of impact measure was introduced recently, called the "audience factor" (AF). It is a variant of the journal impact factor where emitted citations are weighted inversely to the propensity to cite of the source. In the initial design, propensity was calculated using the average length of bibliography at the source level with two options: a journal-level average or a field-level average. This citing-side normalization controls for propensity to cite, the main determinant of impact factor variability across fields. The AF maintains the variability due to exports-imports of citations across field and to growth differences. It does not account for influence chains, as in powerful approaches taken in the wake of Pinski-Narin's influence weights. Here we introduce a robust variant of the audience factor, trying to combine the respective advantages of the two options for calculating bibliography lengths: the classification-free scheme when the biblio...