A variant is a modified version of a page, component, or experience that is shown to a subset of visitors during an experiment. It's the "B" in an A/B test — the version with your proposed change.
The original, unchanged version is called the control (or variant A). Everything else — whether it's a different headline, a new layout, or an entirely redesigned page — is a variant.
Control vs. Variant
| Control | Variant | |
|---|---|---|
| What it is | The current/original version | A modified version with changes |
| Purpose | Serves as the baseline for comparison | Tests whether a change improves performance |
| Also called | Original, baseline, variant A | Challenger, treatment, variant B/C/D |
Single vs. Multiple Variants
In a standard A/B test, there's one control and one variant. But experiments can include multiple variants:
- A/B test — Control + 1 variant (2 versions total)
- A/B/C test — Control + 2 variants (3 versions total)
- Multivariate test — Control + many variants testing combinations of changes
More variants require more traffic to reach statistical significance, since each variant needs enough visitors to produce reliable data.
What Makes a Good Variant
- Tests a clear hypothesis — The change should be tied to a specific reason you believe it will improve performance
- Changes one thing (ideally) — If you change the headline, CTA, and layout all at once, you won't know which change drove the result
- Is meaningfully different — Testing "Buy Now" vs. "Buy Now!" won't produce detectable results. Test substantively different approaches.
Variants in Bandit Testing
In multi-armed bandit testing, multiple variants compete simultaneously and the algorithm dynamically shifts traffic toward the best performers. Variants that underperform get less traffic over time, while winners get more — reducing the cost of showing losing variants to your visitors.