App Store Search conversion fell 36.6% in under two months while impressions tripled

[simonoppowa]

Found while capturing the Apple acquisition baseline for #1078 on 2026-09-08. Filed separately because it is not a listing-map question — it is larger than anything that map could move, and it wants its own answer.

The observation

Two matched 28-day windows from App Store Connect App Analytics, grouped by Source Type. Both windows end before 2.2.0 shipped on 2026-09-04, so the version is not involved.

	Jun 12 – Jul 9	Aug 7 – Sep 3	Change
App Store Search — impressions/day (unique devices)	96	289	3.0x
App Store Search — conversion	39.33%	24.94%	−14.39 pts / −36.6% rel
App Store Browse — conversion	13.25%	8.33%	−4.92 pts
App Referrer — conversion	63.55%	75.20%	+11.65 pts
Web Referrer — conversion	54.25%	61.63%	+7.38 pts

App Store Search is ~78% of all impressions, so this is the dominant discovery surface, not a corner of the funnel.

What is ruled out

• Traffic mix. Decomposing the blended decline: mix effect +0.17 points, rate effect −10.56 points — mix explains 1.6% of the move. The mix slightly improved over the period (Search rose 76.2% → 78.5% of impressions; App Store Browse, the worst-converting source, fell 11.1% → 9.8%).
• The 2.2.0 release. Both windows precede it entirely.
• A general downturn. Both referrer sources moved the other way over the same period, and downloads rose in absolute terms. This is specific to the App Store surfaces.

What is not ruled out

The shape — impressions tripling while conversion falls by a third — is what you would expect if the app began ranking for broader, lower-intent queries. More people see the listing; a smaller share of them wanted this app. Under that reading nothing is wrong with the product or the listing, and the absolute download growth is real.

Other candidates nobody has excluded:

• A change in what the listing shows for those queries — note the live iPhone screenshots are still debug-build captures at 424x900 with a red DEBUG ribbon and an app bar reading OpenNutriTracker [Alpha] (#1082). If broader queries are sending less committed traffic, a listing that looks unfinished converts them even worse.
• Increased competition on the queries that were already converting.
• An Apple-side ranking or surfacing change.
• Something seasonal, though a 36.6% relative move over eight weeks is large for that.

Why it cannot be diagnosed with what Apple gives us

Distinguishing "broader queries" from "same queries converting worse" needs per-query impressions and conversions. App Store Connect does not expose search terms — confirmed on #1078, the Source Type dimension is as granular as acquisition data gets, and Apple’s search-term data is an Apple Search Ads product covering only terms that triggered your own ads.

So the single dataset that would settle the largest movement in this app’s acquisition funnel is the one the platform withholds. Worth stating plainly rather than rediscovering.

Possible next steps, none of them obviously right

• Accept and monitor. Downloads are up in absolute terms; if this is broader reach, the rate falling is arithmetic, not damage. Re-read in a month and see whether it stabilises.
• Fix the listing anyway. #1082 and map #1062 already replace the debug screenshots and the copy. That is worth doing regardless, and it is the one lever available without per-query data.
• A third-party ASO tool would supply keyword-ranking history that Apple does not. That is a paid dependency and an evidence-quality downgrade — vendor estimates, not first-party data — so it should be a deliberate decision, not a reflex.

Consequence for #1078, already recorded there

This drop sets the baseline that the listing change will be measured against, and it is falling steeply for reasons that predate the change. A real improvement would be understated, possibly to zero; a continued decline would be misattributed to the new listing. The pre-window is pinned to Aug 7 – Sep 3 on that ticket so the window cannot be chosen after the fact.

Reproduction

Metrics view grouped by Source Type; the date range is URL-encoded and directly re-readable, but only inside Apple’s 90-day daily-data retention — after roughly 2026-12-06 these exact windows can no longer be read back.

.../analytics/metrics?chartType=singleaxis&dateSpec=r:20260807:20260903&frequency=day&groupByDimensionKey=source&measureKey=impressionsTotalUnique

Swap measureKey=conversionRate for the rates, and dateSpec=r:20260612:20260709 for the early window.

Aggregation note: these are Apple’s daily averages rather than ratios of totals. The decomposition compares like with like throughout, so the conclusion holds under either convention.

Priority untriaged — labelled ios only.