rankcusp explains: the latest changes in the App Store search algorithm

What we do not know

Apple does not publish the ranking algorithm and does not announce its changes. You can only notice one by seeing many apps move similarly on the same day - this document is about how to look.

Most writing on this subject describes something unknown as though it were known. Let us be clear from the start: the App Store's ranking algorithm has not been published.

Apple says in broad terms which fields are indexed, but not how heavily. That the app name is stronger than the subtitle can be observed; how many times stronger is unknown.

That rating score and count affect rank can be observed. At what thresholds and along what curve is unknown.

Download velocity and session length are thought to matter, but that cannot be verified from outside, because neither download nor usage data is available to third parties.

So identifying an "algorithm update" tells you nothing about the algorithm itself. It only says this: something changed, and the effect looks like this.

That distinction matters, because a text saying "Apple made this change, so now you should do that" has turned an observation into an explanation - and the explanation is invention.

Four things mistaken for a change

"The algorithm updated" is an easy explanation to reach for and it is usually wrong. Four alternatives have to be eliminated first, every time.

The first is your own move. If metadata was touched in the last two weeks, that is most likely the cause. The index takes a few days to settle and the effect beds in over a week.

The second is a rating change. If the score fell after a release, your strength weakens across every phrase at once. That produces broad, persistent movement - which looks exactly like an algorithm change.

The third is seasonality. If a group of phrases gains popularity in the same period, the number of apps entering them rises too and holding the same rank gets harder. Comparing against the same period last year separates that.

The fourth is catalogue movement. A large app being pulled, renamed, or a new large app entering the category produces simultaneous shifts across many phrases.

Any "the algorithm changed" finding made without eliminating those four has skipped the most likely explanation and chosen the least likely.

Behaviours that can be observed

The algorithm is unknown, but some behaviours are observed repeatedly and can be verified. These are not a formula, only consistent observations.

The index does not update immediately. A metadata change usually takes a few days to reach search results, and the delay varies by country. Anyone can verify that in their own app.

Repeating the same word in different fields gains nothing extra. It can be tested by removing a word from both the app name and the keyword field and leaving it in only one.

Singular and plural are not kept apart. It can be verified by writing only the singular and observing that you rank for the plural search.

Words in the keyword field combine with one another to form phrases. It can be verified by writing two words separately and observing that you rank for the combined phrase.

The long description is not indexed for search. It can be verified by putting a phrase only in the description and observing that you do not appear for it.

Any claim beyond those five - how many times heavier a field is, what a given threshold does - cannot be verified and is not in this document.

Announced and unannounced changes

Apple announces some changes and not most. Separating the two tells you which sources to trust.

Announced changes are usually about the developer-facing fields: character limits, new metadata fields, publishing rules. Those are written in Apple's own developer documentation and can be verified. If a source mentions that kind of change, you can check it against the document.

Unannounced changes are on the ranking side: weights being tuned, index refresh frequency, how results are arranged. There is no official source about those and there will not be.

Every text discussing that second group is based on observation. The good ones say it is an observation; the bad ones present it as an explanation.

The practical rule: if a claim can be verified in Apple's documentation it is certain; if it cannot, it is an observation. Observations are valuable, but they should be passed on as observations.

This document is an observation document too. The method it describes helps you detect whether a change happened - not explain what it was.

Frequently asked

Does Apple announce algorithm changes?

It announces changes to developer-facing fields and rules; it does not announce weight adjustments on the ranking side. Every text about the second group is based on observation.

My ranks dropped sharply in a day - was it the algorithm?

Probably not. Eliminate the four alternatives first: your own move in the last two weeks, a rating drop after a release, seasonality, and a large movement in the catalogue. Without eliminating them, the algorithm explanation is the least likely one.

I detected a change - what should I do immediately?

Nothing in the first week. The movement usually does not settle and partly comes back. A move made then is ineffective and pollutes the measurement. Mark the record and measure again a week later.

How do I protect myself from changes?

Not by predicting them but by building durably: relevance first, no dependence on one phrase, spread across countries, keep a record, and do not over-optimise. The most tightly squeezed metadata loses the most when a weight moves.

Do I need a special tool for this detection?

You need daily rank history for a few apps. You can keep that by hand; you only have to search the same phrases, in the same country, at the same time every day. The difficulty is not the method, it is being regular.