01The finding
| Storefront | n | Spearman ρ | 95% CI | No suggestions | At the cap (≥10) |
|---|---|---|---|---|---|
| United States (US) | 200 | 0.16 | 0.02 … 0.29 | 0% | 83.5% |
| United Kingdom (GB) | 200 | 0.27 | 0.14 … 0.39 | 1.5% | 64.5% |
| Germany (DE) | 200 | 0.30 | 0.17 … 0.42 | 0% | 62% |
| Spain (ES) | 200 | 0.36 | 0.23 … 0.48 | 1.5% | 56% |
| Turkey (TR) | 200 | 0.25 | 0.12 … 0.36 | 0% | 65.5% |
| All | 1000 | 0.30 | 0.25 … 0.36 | 0.6% | 66.3% |
Across all storefronts the Spearman rank correlation between Apple's search popularity (1–100) and the number of autocomplete suggestions is ρ = 0.30 (n = 1000, 95% bootstrap interval 0.25 … 0.36). The interval excludes zero: the relationship is real and positive, but weak to moderate. Per storefront it ranges from 0.16 to 0.36.
0.6% of these popular terms return no autocomplete suggestion at all, and 99% appear among their own suggestions.
02Why the correlation is what it is
- Ceiling. Suggestion count saturates. Among head terms it says "there is demand", not "how much".
- Restricted range. The sample only holds Apple's head terms: popularity runs from 56 to 100 (median 74). Correlations shrink when one variable barely varies — the low end, where the two signals would disagree most, isn't in the data.
- Different weeks. Apple's popularity data lags by one to two weeks (week 2026-08-09); autocomplete was read on 2026-09-14.
03The floor of 5, and the autumn-2025 collapse
Apple's popularity score runs from 5 to 100, with 5 as the floor. In a four-day window starting 29 September 2025 the number of US keywords scoring above that floor fell from 165,875 to 39,254 (−77.4%, measured by aso.dev). Since October 2025 the API only scores terms above the threshold, and Apple added a monthly Search Term Rank report in beta.
None of the 1000 terms sits at the floor of 5 — the lowest popularity in the sample is 56. Removing floor values therefore changes nothing (ρ stays 0.30), and the question "does clustering at 5 distort the correlation?" cannot be answered from this corpus. We say so instead of inventing a comparison.
That is a direct consequence of the collapse: the corpus Apple now exposes contains only terms with meaningful search traffic. A comparison against the long tail — where most indie keywords live and where a floor value of 5 used to hide real differences — is not possible with Apple's data any more. That gap is the reason autocomplete is the demand signal in Storelift at all.
04Method and limits
- Popularity: Apple Ads Insights search-term popularity, country-wide lists (all genres), top 200 terms per storefront by in-genre rank, week 2026-08-09.
- Demand: number of App Store autocomplete suggestions for the exact term, read on 2026-09-14 with the same request the product uses. 0 of 1000 requests failed and are excluded, not counted as zero.
- Statistic: Spearman rank correlation with average ranks for ties; percentile bootstrap, 2,000 resamples, fixed seed — the interval is reproducible.
- Storefronts: US, GB, DE, ES, TR. Apple Ads popularity only exists where Apple Ads runs.
Frequently asked
So is autocomplete a bad demand signal?
It answers a different question. It separates terms people type from terms nobody types, in every storefront, including the ones where Apple Ads doesn't run. It does not rank the most popular terms against each other — this measurement shows that plainly.
Why not just use Apple's popularity score?
Since the autumn-2025 change it only scores terms above a traffic threshold and only in Apple Ads countries. For most small-app keywords it returns the floor or nothing. Where it exists it is useful; it can't be the only signal.
Can I re-run this?
Yes. The CSV on this page has every row, and the correlation can be recomputed from it with any statistics tool.
See this for your own app
Paste your store link, add your keywords. Rank, the apps above you and demand are measured every day.
Start free →