Lognormal statistical distributions are observed in a variety of scientific fields. The widths of these distributions in the log scale are often similar, but the underlying mechanism that maintains these widths within a small range has not been well explained. We show that a stochastic process of halving followed by addition can yield a stationary distribution that resembles the universal lognormal distribution with a certain width. The mechanism that we propose here would provide insight into the essence of why lognormal-like distributions in many systems have a common width.