A/B testing at the link layer

The step between the ad and the store is testable, and almost nobody tests it.

6 August 2026 · PAD team

What is worth testing

• How long to wait before showing the manual fallback button
• The wording on that button, which changes its use rate substantially
• Whether an interstitial with context beats an immediate redirect
• Sending straight to the store versus a landing page first

The first two are the highest-leverage and the least examined. In our own traffic the button's copy moves its use rate more than most creative changes move click-through.

Splitting cleanly

Split on a stable hash of the visitor rather than randomly per request, so someone who taps twice sees the same variant. Otherwise your measurement includes people who experienced both.

const bucket = hash(visitorToken) % 2 === 0 ? 'A' : 'B';

Keeping attribution intact

Every variant must attach the same campaign token, or your test contaminates campaign reporting. Encode the variant separately in your own analytics rather than in ct, otherwise App Store Connect fills with duplicated campaign names.

What to measure

Not installs. The link layer affects arrival rate, and installs depend on your store page too. Measure the share of taps that reached the store, which isolates the thing you changed.

Sample size

Arrival rates differ enough between variants that differences show up faster than in creative testing, but the traffic is also segmented by platform. Run the test long enough that each platform has meaningful volume, because a change that helps Facebook may do nothing for Instagram.