Abstract The circumgalactic medium (CGM) close to ∼L* star-forming galaxies hosts strong Mg ii λ2796 absorption (with equivalent width W 2796 > 0.1 Å) with a near-unity covering fraction. To characterize the spatial coherence of this absorption, we analyze the W 2796 distribution in the CGM of 27 star-forming galaxies detected in deep spectroscopy of bright background (b/g) galaxies first presented in Rubin et al. The sample foreground (f/g) systems have redshifts and stellar masses 9.1 < log M */M ⊙ < 11.1, and the b/g galaxies provide spatially extended probes with half-light radii at projected distances < 50 kpc. Our analysis also draws on literature W 2796 values measured in b/g QSO spectroscopy probing the halos of f/g galaxies with a similar range in M * at z ∼ 0.25. By making the assumptions that (1) samples of like galaxies exhibit similar circumgalactic W 2796 distributions and, (2) within a given halo, the quantity log W 2796 has a Gaussian distribution with a dispersion that is constant with M * and , we use this QSO–galaxy pair sample to construct a model for the log W 2796 distribution in the CGM of low-redshift galaxies. Adopting this model, we then demonstrate the dependence of the observed log W 2796 distribution on the ratio of the surface area of the b/g probe to the projected absorber surface area (x A ≡ A G/A A), finding that distributions that assume x A ≥ 15 are statistically inconsistent with that observed toward our b/g galaxy sample at a 95% confidence level. This limit, in combination with the b/g galaxy sizes, requires that the length scale over which W 2796 does not vary (i.e., the “coherence scale” of Mg ii absorption) is ℓ A > kpc. This novel constraint on the morphology of cool, photoionized structures in the inner CGM suggests that either these structures each extend over kiloparsec scales or the number and velocity dispersion of these structures are spatially correlated over the same scales.