How to A/B test links with one short URL

The short link is the allocation layer: each open is routed according to the destination weights. This differs from changing one button inside a page. It also has an important limitation: taap.at assigns a destination on each open rather than keeping one variant for each person, so a repeat visitor can encounter A and later B.

What one short-link test can answer

Prepare a test that can produce an answer

1. 1

   Write the hypothesis

   Name the audience, change, and expected outcome: for example, showing the price will increase sign-up rate compared with the current page.

2. 2

   Choose one primary outcome

   Use a business result that both destinations can record. Add guardrails such as refunds, abandonment, or lead quality when an apparent lift could hide a cost.

3. 3

   Build the control and variant

   Keep everything else stable when you need to isolate a change. A whole-page test can identify the better package, but it cannot tell you which component caused the result.

4. 4

   Set allocation, sample, and stopping rule

   A 50/50 split usually makes efficient use of traffic. Estimate sample size from baseline conversion, the minimum effect worth acting on, and your statistical method; there is no universal click or day threshold.

5. 5

   QA routing and analytics

   Open the link repeatedly, check both destinations on mobile and desktop, and confirm every conversion retains its variant label before sending campaign traffic.

Create the link and preserve attribution

Generate the short link, then sign in to add the second destination and assign weights. Percentage-based traffic splitting is available on every plan. A 50/50 allocation is a probability, so observed counts do not need to be exactly equal in a finite sample.

Keep utm_source, utm_medium, and utm_campaign consistent to preserve acquisition context, then distinguish A and B with utm_content or another non-personal label. A UTM identifies the variant; GA4, your CRM, checkout, or sales system must record the actual outcome.

Analyze without manufacturing an early winner

• Do not change pages, weights, the primary metric, or tracking mid-test; start a new phase after a material change.
• Do not stop at the first favorable spike. Predefine a sample and stopping rule compatible with your method.
• Cover relevant business cycles, such as weekdays and weekends, without treating seven days as a universal rule.
• Check whether bots, preview crawlers, or security scanners inflated opens; a recorded click is not necessarily a person.
• If the uncertainty still includes worthwhile gains and harmful losses, record an inconclusive result rather than calling it a tie.
• Investigate a persistent mismatch between expected and observed allocation before trusting the result.

Know the limits of per-open allocation

Because taap.at assigns on every open, this setup is best for destination comparisons where exposures can be analyzed independently. It is a poor fit when a person must retain the same experience across days, prices must remain consistent, or conversion requires several visits. Use an experimentation platform with persistent user assignment and integrated statistics for those cases.

SEO, privacy, and ending the experiment

When variants are indexable versions of the same page, Google recommends avoiding cloaking, pointing canonical tags to the preferred URL, using temporary redirects where appropriate, and removing the experiment when it ends. If you use cookies or persistent identifiers, assess notice and consent for each market, and never place personal data in URL parameters.

What to do with low traffic

If the required sample would take months, do not force a decision. Customer interviews, usability sessions, and reviews of calls or forms can expose larger problems faster; you can also test a more meaningful difference. Record an inconclusive result and keep the current experience until evidence supports a change.