How long does it take for ASO changes to affect rankings?

We ran the obvious measurement on our own corpus: 888 changes to an indexed text field, and how long until the tracked keyword moved. Then we ran the control, and the control came back the same. Here is what that means for reading your own release.

Search for this question and you will find confident answers: 24 hours, 48 hours, one week. The problem is that those numbers usually collapse several different events into one. Submission is not release. Release is not the moment every storefront displays the new listing. Detection is not the moment the change occurred. A ranking observation is not proof that the edit caused it.

The honest answer is a measurement design. Once the timestamps are separated, you can report what happened to your app without pretending the store runs on a public schedule.

We ran the measurement

The corpus makes the obvious test possible. We record a daily position for every tracked keyword and we detect changes to tracked listings, so we can ask directly: after an app changes text the store says it indexes, how long until its keywords move?

Restricting to the fields the stores name as indexed \u2014 name, subtitle and description \u2014 gives 888 changes. Pairing each with the app\u2019s tracked keywords, and requiring at least five readings in the fortnight before so there is a baseline to compare against, gives 1,112 testable keyword and change pairs. A \u201cmove\u201d means the first reading that falls outside the range of that keyword\u2019s prior fortnight. Keywords already pinned at position one are excluded, because they have no visible upside and would dilute the result.

48% of those pairs moved, at a median of 2.1 days. A quarter moved within a day, and three quarters within five. Read on its own, that looks like the answer the question wants: about two days.

Then we ran the control, and it matched

That number is worthless without knowing how often a keyword breaks its own fortnight range anyway. Ranks move on their own every day. So we repeated the identical test anchored on random dates for the same keywords, excluding any date within a fortnight of a real change: 7,736 keyword and date pairs where, as far as we can tell, nothing was edited at all.

The control is indistinguishable from the treatment, and on the headline figure it is slightly higher. Roughly half of all tracked keywords leave their two-week range within a couple of days regardless of whether anybody touched the listing. The \u201ctwo days\u201d finding was a description of App Store noise, not of indexing.

We also checked direction, since a test that counts a fall as a hit is blind to whether the edit helped. Comparing mean position in the fortnight before against the fortnight after, keywords after a change drifted 1.53 positions worse on average, against 0.63 worse on placebo dates; 28.8% improved, against 35.9% on placebo dates.

What this does and does not establish

It does not establish that metadata changes have no effect. A test this crude would miss a real effect of a few positions, the post-change windows are short because our collection began in July 2026, and 1,112 treated windows is not a large sample. Most of these changes are also bundled: a subtitle rewrite usually ships with a new version, new screenshots and a new icon on the same day.

What it does establish is narrower and more useful. The specific inference almost everybody makes \u2014 I changed my subtitle, my rank moved three days later, therefore the subtitle worked \u2014 does not survive its own control. If you want to know whether an edit worked, the movement of a few keywords in the following week is not the instrument.

The three clocks

Only the owner reliably knows the first timestamp. An outside observer usually knows the second as an interval: the old listing was present at one reading and the new listing was present at the next. A daily collector therefore knows the change happened somewhere inside that window, not at the exact instant of the second reading.

A real observation window

In AsoTheory’s early corpus, Duolingo’s iOS version timestamp showed a release on 27 July 2026 at 16:19 UTC. The US listing was last observed in its earlier state on 26 July at 11:20 and the new version, icon and size were detected on 29 July at 20:58. Without the store’s version timestamp, the honest external statement would be “the change occurred inside an 81-hour observation window.” The version timestamp narrows one part of that event, but it does not provide an exact icon-change time.

The tracked US rankings did not deliver a simple success story. “Learn spanish” remained first on iOS before and after the release. “Language learning” also remained first on iOS. On Android, “language learning” moved from first to second and back to first while “learn spanish” stayed first. That is not evidence that the release helped, harmed or had no effect: the listing did not make a clean, isolated keyword change and the terms were already at the ceiling.

What the stores actually say

Apple identifies the title, subtitle, keywords and primary category as text-relevance inputs, alongside behavioral factors including downloads, ratings and reviews. Apple allows up to 30 characters for the name and subtitle and up to 100 bytes for the keyword field. Google Play exposes a 30-character app name, an 80-character short description and a 4,000-character full description, and says ranking also incorporates relevance, app quality and user response.

Those documents describe eligible inputs, not a fixed reindexing delay. Any article promising one universal response time is adding certainty the platform owners do not publish.

How to run a clean metadata measurement

The answer you can safely publish

For your own app, report a range: “The new subtitle became visible between Tuesday’s and Wednesday’s observations; the target term moved outside its four-week baseline two days later and held for seven days.” That sentence is more useful than “App Store changes take 48 hours,” because every part of it can be checked.

For a competitor, be even stricter: “We detected the change on Wednesday; it occurred after Tuesday’s observation.” Unless the store exposes a release timestamp for the changed field, the detected time is not the edit time.

Blog · Case studies · Free ASO tools · Products · Privacy and terms · AsoTheory