Split testing runs two or more variants with equal conditions to isolate what drives lift. Platform native tools (Meta Experiments, Google Experiments) handle traffic split; for clean tests change one variable, run long enough for conversions, and pre-register success metrics.
Avoid editing both variants mid-test — invalidates results.
Sources
Last verified Jun 5, 2026