An A/B test compares two versions of something to learn which works better for a specific goal. For example, some visitors may see sign-up page A and others see page B.
To make the comparison fair, the two groups should be assigned in a planned way, the same measure should be used for both, and the team should decide ahead of time when to review the result.
Seeing one version briefly lead does not prove it is better. A change in traffic, tracking, or timing can affect the numbers, so the test needs a clear plan and careful reading.