Short definition
A/B testing is splitting traffic between two versions of a page or element to measure which one produces a better outcome with statistical confidence.
In depth
A/B testing replaces opinion with evidence. You build a control (A), a variant (B), randomly assign visitors, and measure a single primary metric (conversion rate, revenue per visitor) until results reach significance. Common tests: headline, hero image, CTA copy, form length, social proof placement. For sites with low traffic (<5,000 visits/month) sequential design or multi-armed bandit approaches outperform classic A/B due to sample-size limits.
Example
A web design agency tests 'Get a Free Audit' vs 'Get a Custom Proposal' and finds proposal copy converts 18% better. The site adopts the winner.