Metaphors such as “rabbit hole” or “intellectual black hole” suggest that endorsing conspiracy theories makes people impervious to evidence challenging their views. Yet, this claim remains untested. Building on the Bayesian approach to belief update, we first provide a framework to distinguish confirmation bias in evidence evaluation (i.e., the general tendency to distrust evidence challenging prior beliefs and attitudes) from conspiracist evidence discounting (i.e., evidence discounting attributable to the specificities of conspiracy theories). We propose that the former results from the strength of prior beliefs, while the latter is a consequence of the structure of prior beliefs – and specifically, the fact that conspiracy theories entail a deception assumption that preemptively defuses disconfirming evidence. We then experimentally test hypotheses derived from this framework.