Central banks around the world have revised their operating frameworks in an attempt to counter the challenges presented by the effective lower bound (ELB) on policy rates. We examine how private sector agents might learn such a new regime and the effect of future shocks on that process. In our model agents use Bayesian updating to learn the parameters of an asymmetric average inflation targeting rule that is adopted while at the ELB. Little can be discovered until the economy improves enough that rates would be near liftoff under the old policy regime; learning then proceeds until either the new parameters are learned or the average inflation target is reached. Recessionary shocks forcing a return to the ELB would thus delay learning while large inflationary shocks could outright stop it and so inhibit the ability of the new rule to address future ELB episodes. We show the central bank can offset some of the inflation-induced learning loss by deviating from its new rule, but it must weigh the benefits of doing so against the costs of higher near-term inflation and greater uncertainty about the policy function.