Each study draws n values from a population and records their average x̄. Repeat the study many times. The middle panel shows the averages on the original scale: as n grows they gather at the true mean μ, and the share outside the band μ ± σ/2 falls to zero (the law of large numbers). The bottom panel standardises them, Z = (x̄ − μ)/(σ/√n): whatever the shape of the population, the histogram approaches the gold bell curve (the central limit theorem).
Two different statements: the law of large numbers says x̄ ends up close to μ; the central limit theorem describes the shape of its random error on the scale σ/√n. The second is what makes ± 1.96 standard errors work for non-normal data.