Definition
A/B testing is a controlled experiment method: users or traffic are randomly split into groups that experience different variants (A and B), and a metric chosen in advance decides which variant performs better.
How it works
The keys are randomization and controlled variables: random assignment keeps the groups comparable on everything else, and changing only the element under test (such as a headline or button copy) lets any metric difference be attributed to that change. Sufficient sample size and run time are also required before conclusions are reliable.
Usage
A/B testing is widely used for web and interface optimization, growth experiments and recommendation evaluation — for example, comparing signup conversion between two versions of subscription page copy.
Related and confusable terms
A/B testing differs from benchmarking: the former compares changes against one goal, while the latter measures overall system capability on fixed tasks. Verdicts in A/B tests usually rest on key performance indicators (KPIs) such as conversion or click-through rates.