The short GRB 170817A associated with the first detection of gravitation waves from a binary neutron star (BNS) merger was in many ways unusual. Possible explanations are emission from a cocoon or cocoon break out, off-axis view of a structured or uniform jet, and on-axis ultra-relativistic jet with reduced density and Lorentz factor. Here, we use a phenomenological model of shock evolution and synchrotron/self-Compton emission to simulate the prompt emission of GRB 170817A and to test above proposals. We find that synchrotron emission from a mildly relativistic cocoon with a Lorentz factor of 2–3, as considered in the literature, generates a too soft, too long, and too bright prompt emission. Off-axis view of a structured jet with a Lorentz factor of about 10 can reproduce observations, but needs a very efficient transfer of kinetic energy to electrons in internal shocks, which is disfavoured by particle in cell simulations. We also comment on cocoon breakout as a mechanism for generation of the prompt gamma-ray. A relativistic jet with a Lorentz factor of about 100 and a density lower than typical short GRBs seems to be the most plausible model and we conclude that GRB 170817A was intrinsically faint. Based on this result and findings of relativistic magnetohydrodynamics simulations of BNS merger in the literature we discuss physical and astronomical conditions, which may lead to such faint short GRBs. We identify small mass difference of progenitor neutron stars, their old age and reduced magnetic field, and anti-alignment of spin–orbit angular momentum induced by gravitational disturbances during the lifetime of the BNS as causes for the faintness of GRB 170817A. We predict that BNS mergers at lower redshifts generate on average fainter GRBs.