Both indicative and counterfactual conditionals are known to be licensing contexts for negative polarity items (NPIs). However, a recent theoretical account suggests that the licensing of attenuating NPIs like English all that in the conditional antecedent is sensitive to pragmatic differences between various types of conditionals. We conducted three behavioral experiments in order to test key predictions made by that proposal. In Experiment 1, we tested hypothetical indicative and counterfactual conditionals with the English NPI all that, finding that the NPI is degraded in the former compared to the latter. In Experiment 2, we compared hypothetical indicative conditionals and premise conditionals with the same NPI, again finding a degradation only for the former. Both results align with theoretically derived predictions purporting that hypothetical indicative conditionals are degraded due to their susceptibility to conditional perfection. Finally, Experiment 3 provides empirical evidence that comprehenders readily strengthen counterfactual conditionals to biconditionals, in line with theoretical analyses that assume that conditional perfection and counterfactual inferences are compatible. Their ability to still host attenuating NPIs in the conditional antecedent, by contrast, falls into place via the antiveridical inference to the falsity of the antecedent. Altogether, our study sheds light on the interplay between NPI licensing and the semantic and pragmatic properties of various types of conditionals. Moreover, it provides a novel perspective on the processing of different kinds of conditionals in context, in particular, with regard to their (non)veridicality properties.