Running your first A/B test

A/B testing works best on links that already receive meaningful traffic. The most common rookie mistake is running a test on a link with 10 clicks a week and stopping after 4 days — you'll see noise and call it a result.

Statistical reliability requires at least 100 clicks per variant. With a 50/50 traffic split, that means 200 total clicks before you can even start interpreting numbers. If your link gets 50 clicks a week, that's about 4 weeks per test — and that's if you wait properly.

How to find your testing candidate

Go to your link manager. Sort by weekly clicks. The top 1–3 links are your candidates. Pick whichever one has the most strategic upside — usually your hero offer, your top product, or your main CTA.

What to test, in order of impact

• Link title — the most impactful, easiest to iterate on. Change a "Watch my latest video" to "How I record vocals at 3am" and you might double CTR overnight.
• Cover image / thumbnail — second-biggest lever. Tested less often because creators worry about brand consistency.
• Description text (subtitle below the title) — meaningful but smaller swings.
• Link order on the page — affects every link, not just one. Run carefully and only on big audiences.
• Button colour or shape — almost always within noise for normal-sized audiences. Don't bother unless you have 10,000+ weekly visitors.

Click the three-dot menu on your chosen link and select "Create A/B test". The current link becomes Variant A automatically — that's your control. You'll then enter the alternative version (Variant B), and that's your hypothesis.

1. 1Enter the Variant B title — the alternative you want to test
2. 2Set the traffic split (default 50/50, recommended)
3. 3Optionally add a hypothesis note — a sentence describing why you think B might win
4. 4Click "Start test"

Writing a strong Variant B

If your two variants are too similar, the test will produce a tie no matter how long you wait. Make Variant B meaningfully different from A. Aim for 'these are clearly different framings' rather than 'one has a comma the other doesn't'.

Examples of meaningful differences:

• A: 'Watch my latest YouTube video' B: 'Behind the scenes of my album recording' (descriptive vs curiosity-driven)
• A: 'Get my course' B: 'Join 800+ photographers learning Lightroom' (generic vs social proof)
• A: 'Free preset pack' B: 'The presets I use on every paid shoot' (offer vs personal)
• A: 'Buy now' B: 'Drop closes Friday' (cta vs urgency)

Variant A: "Watch my latest YouTube video"
Variant B: "Behind the scenes of my album recording 🎵"

Hypothesis: Specific, curiosity-driven titles outperform generic
            ones for our music audience.

Once started, the test runs automatically. Your job now is to wait. Statistical significance requires a minimum sample size — and waiting through a full week captures normal day-of-week variation.

What "enough data" actually means

• Minimum recommended runtime: 7 days. Always. Even if you're 95% confident on day three.
• Minimum sample: 100 clicks per variant before reading results — 200 if you can manage it.
• Confidence target: 95% or higher before declaring a winner. Below 90% is essentially a coin flip.
• If neither variant reaches significance after 30 days, declare the test inconclusive — the variants perform equally for your audience.

Why a full week matters

Visitor behaviour varies systematically by day of the week. Saturday-afternoon traffic is qualitatively different from Wednesday-lunch traffic. A test that runs Sunday-Wednesday over-weights weekend behaviour and under-weights weekday. One full week (Monday-to-Monday or any 7-day stretch) gives you a representative sample.

Go to Analytics → A/B Tests to view your running and completed experiments. Each test card shows clicks, impressions, click-through rate, uplift, and confidence.

Understanding the metrics

• Impressions — how many visitors saw each variant. With 50/50 split, these should be roughly equal.
• Clicks — total clicks per variant. Useful as a sanity check; not the metric you optimise.
• CTR — click-through rate (clicks ÷ impressions). THIS is the metric you care about.
• Uplift — percentage improvement of B over A in CTR terms. +20% means B converts 20% better than A.
• Confidence — statistical certainty of the result. Aim for 95%+ before declaring a winner. Below 90% is too noisy to trust.

Worked example

Variant A got 150 clicks from 400 visitors (37.5% CTR). Variant B got 120 clicks from 350 visitors (34.3% CTR). Variant A wins — even though B had fewer raw clicks. The traffic split was slightly uneven, and per-visitor performance is what matters.

If Variant B reaches 95% confidence with a positive uplift, click 'Declare winner'. LinkStacked replaces the original link with the winning version and archives the test data for future reference.

The single test that ran in isolation isn't the real value of A/B testing. The real value is the pattern that emerges over months of testing — a body of evidence about what works for your specific audience.

Document everything

After every completed test, write down: what you tested, what won, by how much, and your one-line interpretation of why. After 5–10 tests, patterns emerge. Maybe your audience always responds to specific over generic. Maybe urgency consistently wins. Maybe social proof barely matters. You'll never see this pattern from a single test, but you can't miss it after a quarter of testing.

Pick the next test

After declaring a winner, immediately queue the next test. The hardest part of an A/B testing programme is keeping it going — most creators run their first test, get a result, and never run another. Calendar your next test before closing the dashboard.