A/B testing starts by randomly dividing the website’s traffic into two groups. One group sees the original version of the website, while the other group sees a variation that may include a different headline, new design, or other changes. The two versions are then compared based on a specific goal, and the version that performs better is selected as the winner.