A/B Testing

Run A/B tests directly inside the WordPress block editor — no separate tool, no code, no developer required. Create a variant, pick a goal, and Accelerate handles the traffic split, the measurement, and the statistics. You just decide what to try.

What you can test#

  • Blocks. Test any block content — headlines, CTAs, images, layouts — by creating variants right in the editor.
  • Post titles. Try multiple titles for a post and let real readers pick the winner. Works on both block and classic themes.
  • Featured images. Test which featured image earns more clicks from archives, listings, and the homepage.

How results are measured#

Accelerate uses Bayesian statistics to evaluate your tests. In practice that means you don't need to know anything about p-values or sample sizes: the results screen shows each variant's conversion rate and tells you how confident you can be in the difference. When a result is ready, calling the winner is one click.

How we know it really won

Phase 3 of 4: Confident
Each curve is how sure we are of a version's true conversion rate. Tall and narrow means confident, wide and flat means we're still guessing. Watch them sharpen and pull apart as data comes in: while they overlap the result could still flip, but once the purple variant curve clears the gray one, the win is real and not luck.

Performance updates in real time, so you can watch a test from the moment it goes live.

Traffic that follows the winner#

New tests run as multi-armed bandits: instead of holding a fixed split for the whole test, Accelerate shifts more traffic toward the variant that's pulling ahead while the test is still live, so fewer visitors see the losing option. Guardrails keep it honest — every test opens with an even burn-in split, a floor keeps each variant in play, and the split returns to even for a confirmatory phase before a winner is declared. Prefer a classic fixed split? Tests already running keep their behavior, and a filter switches new tests back to even allocation.

Traffic moving to the winner

Phase 2 of 3: Reallocating
Three lines, one per version, showing each one's share of traffic across the test. They start bunched near an even split, then the winner's line (the bright one) climbs toward the top while the two losers sink. It's the moment-by-moment picture of traffic moving onto what works.

Let an agent run the test#

The whole test lifecycle is also available to AI agents through the Abilities API — a tool like the Accelerate AI Toolkit can create a test, start it, read live results at every phase, and apply the winner without anyone touching the editor. Tests carry their origin, so the experiments overview shows at a glance which were created by an agent and which you set up by hand — and a live badge marks any test an autopilot loop is actively managing, so nobody pauses a test the loop is mid-run.

A simple workflow that works#

  1. Pick one measurable goal before you start — a click, a form submission, an engagement.
  2. Create variants that differ in one meaningful way, so you know what caused the result.
  3. Let the test run long enough to collect representative traffic. Resist peeking too early.
  4. Call the winner, keep the learning, and line up the next test.

You don't need a grand experimentation program to benefit. One small test a week compounds quickly — and every test teaches you something about your audience, even the ones that don't produce a winner.