How App Store Ranking Works: Not a Score, a Queue of Signals

Someone asked me last week whether we have access to the latest algorithm. He had heard it changed again last month, and his rankings had dropped.

I said we do not know the algorithm. Nobody does. We know roughly what it looks at, and we adjust for that.

He was disappointed. What he wanted was a password — which parameter moved this time, so he could change it back. That answer does not exist. A lot of people make a living selling it anyway.

The short version

Store ranking is not a score. It is a set of signals, ordered, in a particular context.

You cannot optimise a ranking, because ranking is an output. The only things you can touch are the signals themselves: what your fields say, what users do after they see you, and whether the app holds up over time.

Do those three well and ranking follows. Chase the ranking directly and you will keep getting it wrong, because you cannot see what happened in between.

Why the word "ranking" misleads people

It sounds like standing on a scale and being given a number. Find a way to make the number bigger.

It is closer to hiring. The same candidate applying to two different roles gets two different outcomes. The job description decides who sees them, whether they see the right role, and how the interviewer judges fit. You cannot "increase your hiring probability" directly. You can only make the application match the role.

Search works the same way. The results page for `bookkeeping` and the results page for `simple household bookkeeping` may be led by two different apps. It is not one score computed twice.

Three signal groups: relevance, behaviour, quality

Roughly, and the specific weighting is not public — check the current official guidance.

**Relevance**: how well your fields match the query. This is the only one you control directly.

**Behaviour**: whether people tap your position, install, and keep using it. You control this indirectly, through the icon, the screenshots, the rating, the description — the things that make someone willing to tap.

**Quality**: rating, crash rate, update cadence. This is built over months.

These do not add up. Relevance gets you into the candidate pool. Behaviour decides where you sit inside it. Quality decides whether you deserve to stay.

Relevance is not keyword matching

The common misreading is that relevance means "is the word in my fields".

It has at least two layers. The first is coverage — does the user's phrasing appear anywhere in your metadata. The second is intent — for the same word, are you the thing they actually want.

Take `bookkeeping`. Some searchers want personal finance. Some want small business books. If your fields cover the word but not the intent, you will be included in the results and ranked behind the apps that match intent better.

That is not the algorithm punishing you. That is users choosing. A page with poor relevance gets a low tap rate, and next cycle it drops.

Intent also shifts across markets. The same term in German and in English can describe two different use cases. A translated field covers the word and misses the intent. This is why localisation cannot be solved by translation — translation only handles the first layer.

Behaviour signals cut both ways

Behaviour means the platform watches what happens to the users you receive.

More taps, more installs, longer use — you are valuable. Many impressions, few taps — your position is wrong for you, or your icon and screenshots are not doing their job.

The harsh part is the compounding. Good position leads to more impressions, which leads to more taps, which leads to a better position. The reverse is equally true: once you fall off the first few screens, impressions collapse and you no longer get a chance to prove anything.

So in practice, "improving ranking" means "converting while you still have impressions". Impressions are finite. Optimising after they are gone costs far more.

One note on paid traffic, because it is widely misunderstood. It is not for pushing rankings. Short bursts bring users who generate no real behavioural signal, and if they install and delete, retention falls. What paid traffic is actually good for is accelerating a product that already retains people, so real signals accumulate faster. If the product does not hold users, spend just amplifies a negative.

The quality signal people forget

Everyone watches the rating. One item gets ignored: update stability.

An app that has not shipped in a long time reads as abandoned. Users who see a two-year-old update date are less willing to install. That is not a direct penalty — it shows up through behaviour and conversion.

This is why some teams optimise the metadata carefully and still do not move. The product is scoring negative, and the fields are adding to a negative base.

The two stores do not weight the same things

Worth stating plainly: App Store and Google Play are not the same system.

Broadly, App Store leans more on text relevance and conversion, Google Play leans more on usage behaviour and retention. The same app with the same metadata often performs differently across the two. This is why.

Neither store publishes a complete weighting, and anyone claiming a precise internal percentage table is not credible. Check the current official guidance.

In practice: spend more time on fields and screenshot conversion on iOS, more time on retention and ratings on Android. Copying one set of copy to both usually underperforms on both.

What to change when the algorithm updates

The standard reaction to an update rumour is for everyone to rewrite the fields. That is the worst response.

First, you cannot confirm an update happened rather than a seasonal fluctuation. Rankings move every day. People find causes in noise.

Second, even if it did happen, fields are not the first thing to check. Look at the data: did impressions fall or did conversion fall? All terms or specific ones? New users or existing ones?

Different answers mean completely different work. Falling impressions suggest a relevance problem. Falling conversion suggests a competitor changed their screenshots. One term dropping may just mean competition on that term changed.

Stop benchmarking against a competitor's rank curve

A lot of optimisation is done by watching a competitor: they are at this position, I want that position, they moved up so I panic.

Two problems with that.

You are watching an output, not their action. They may have run a paid burst, been picked up by a feature, changed their main screenshot, or simply had more active users that fortnight. Copying their result rather than their action usually sends you the wrong way.

And their ranking may not share your cause. They may rank on brand volume while you rank on long-tail coverage. Both in the top ten, but copying their approach would cost you your own advantage.

What is worth comparing is not position. It is what their listing promises, what their reviews complain about, and how often they ship. Those you can read. A rank curve you cannot.

When chasing the algorithm is a waste of time

Three cases.

When retention is poor. Users install once and leave, behaviour signals are negative, and no amount of metadata will pull it up. Fix the product.

When the rating is low and the reviews are angry. People arrive, read the reviews, and leave. Conversion is negative. Handle the reviews, not the keywords.

When the category is owned by a dominant player. The problem is not your signals, it is the lane you chose. Change terms or change markets.

The pattern: the bottleneck is not at the SEO layer. Grinding on something that is not the bottleneck is the easiest way to feel productive while nothing moves.

What to do next

1. Open your dashboard and look at impressions and conversion separately. Download count alone never tells you which part is broken. 2. Pick three core terms and search each one. Look at the top results and ask what their screenshots promise that yours do not. Not position — promise. 3. Check when your app last shipped. If it has been more than six months, do that before touching metadata. 4. Record impressions, tap rate and conversion before the change. Check again two weeks later. Without a baseline, all optimisation is a feeling.

---

If a term is stuck and you are not sure whether the problem sits in the fields, the conversion, or the product itself, send us the app and a few target terms. We will tell you which layer the bottleneck is in before discussing whether to work on it. That diagnosis is usually worth more than the execution.