The magnitude–frequency distribution (MFD) of many earthquake catalogs is well described by the Gutenberg–Richter (GR) law or its tapered version (TGR). This distribution is usually extrapolated to any subsets of the space–time window covered by the catalog. However, some empirical observations and logical thoughts may raise doubts about the validity of this extrapolation. For example, according to the elastic rebound theory, we may assert that the probability of a strong shock nucleating within a short‐time interval in a small area just ruptured by another strong event should be lower than that expected by GR (or TGR): a lot of energy has already been released, and it takes time to recover to the previous state. Here, we put forward a space–time modification of the TGR, named energy‐dependent TGR (TGRE) in which the corner seismic moment becomes a time‐varying energy function depending on (1) the conceivable strongest shock that may nucleate in ; (2) the time elapsed since the last strong earthquake that reset the elastic energy in to a residual value; and (3) the rate of the energy recovery, linked to the recurrence time of the fault(s) involved. The model also verifies an invariance condition: for large space–time windows, the occurrence of a strong shock does not affect significantly the whole elastic energy available, that is, the TGRE becomes the TGR. The model is simple and rooted in clearly stated assumptions. To evaluate its reliability and applicability, we apply it to the 1992 Landers sequence. As expected by TGRE, we find that the MFD close to the fault system interested by the mainshock ( 7.3) differs from that of earthquakes off‐fault, showing a lower corner magnitude. We speculate that TGRE may be profitably used in operational earthquake forecasting and that it explains the empirical observation that the strongest aftershocks nucleate always outside the mainshock fault.