An A/B test randomly splits users into a control group (A) and a variant group (B), exposes each to a different experience, and measures the impact on a chosen metric. With enough sample size and a clean random assignment, the difference between the two groups can be attributed to the variant.
Doing it rigorously requires pre-registering the hypothesis, picking a primary metric, computing the required sample size in advance, and not peeking at the result early. Most teams underpower their tests and over-trust early signals.