We investigate information geometry in a toy model of self-organised shear flows, where a bimodal PDF of x with two peaks signifying the formation of mean shear gradients is induced by a finite memory time γ −1 of a stochastic forcing f . We calculate time-dependent Probability Density Functions (PDFs) for different values of the correlation time γ −1 and amplitude D of the stochastic forcing, and identify the parameter space for unimodal and bimodal stationary PDFs. By comparing results with those obtained under the Uniform Coloured Noise Approximation (UCNA) in Jacquet, Kim & Hollerbach (Entropy 20, 613, 2018), we find that UCNA tends to favor the formation of a bimodal PDF of x for given parameter values γ −1 and D. We map out attractor structure associated with unimodal and bimodal PDFs of x by measuring the total information length L ∞ = L(t → ∞) against the location x 0 of a narrow initial PDF of x. Here L(t) represents the total number of statistically different states that a system passes through in time. We examine the validity of the UCNA from the perspective of information change and show how to fine-tune an initial joint PDF of x and f to achieve a better agreement with UCNA results.