Subscribe
App Store OptimizationFacts as of

App Store Optimization 2026: How Search Ranking Works

The short answer

How App Store search ranking works in 2026

Apple says App Store search results depend on text relevance (your title, subtitle, keywords and primary category) and on user behaviour such as downloads, ratings and reviews. Our working model: your words decide whether your app is a candidate for a search at all, then text fit plus taps and downloads on that search decide the order. So give every search you can truly serve its words, and start with specific searches.

Key facts

Chapters

  1. 0:00App Store optimization 2026: what decides search ranking?
  2. 0:43What does Apple say decides search?
  3. 2:20What did Apple's ranking paper test?
  4. 4:05Why does +0.24% matter?
  5. 5:17Does retention move your rank?
  6. 6:46Did my rank drop, or did the page change?
  7. 7:59What decides where your app shows up?
  8. 10:18What can't this tell you?

Transcript

Edited for reading

Coral is a fictional document scanner app; its listing and ranks are illustrative. Every fact about the App Store comes from a real source, listed under Sources.

App Store optimization 2026: what decides search ranking?Watch from 0:00

Apple's researchers tried a new way to teach App Store search ranking, and tested it on worldwide traffic. The result? Plus 0.24%.

That sounds like nothing. But that number, and the paper behind it, show what App Store optimization is really working with. And the biggest gains came on rare, specific searches. That's where small apps get found.

The AI in that paper also does something most people get wrong. By the end, you'll know what Apple has confirmed and what's folklore, and which part of your listing to fix first.

Every claim gets a badge (confirmed, observed or unproven), with facts as of 2 October 2026. Our example app, Coral, is made up. The facts about the store are real.

So what does Apple itself say decides where your app shows up?

Meet Coral. It's a new document scanner, and it just launched. Its listing is a name, a subtitle, and a hidden keyword field. Notice what it says about receipts, by the way. Nothing. Hold on to that.

Apple's own developer page answers the ranking question in one sentence. Search results depend on text relevance: your title, subtitle, keywords and primary category. And on user behaviour: downloads, ratings and reviews, and more.

Apple also says people can search in plain, everyday language. So matching goes beyond exact strings.

Notice the two words at the end: and more. Apple isn't giving you the full list. Notice what's missing, too. Your long description isn't on it. Plenty of guides still tell you to pack keywords into the description. Apple doesn't list it.

Apple says one more thing directly. Ratings and reviews influence how your app ranks in search.

Then there are the limits. Thirty characters for the name, thirty for the subtitle, and a keyword field of 100 bytes that only Apple sees. Your company name is already searchable. And other apps' names aren't allowed anywhere in it.

Why care so much about search? Apple says almost 65% of App Store downloads happen right after one. That figure is Apple's, and it comes from 2022 data. Apple repeated it when it announced more search ads.

Our take: that one sentence from Apple is worth more than most ranking-factor lists. Start there.

What did Apple's ranking paper test?Watch from 2:20

But that sentence is short. Is there anything deeper from Apple? There is. In February 2026, six Apple researchers published a paper on App Store search. Apple rarely says this much about search in public. The latest version came out in June.

The ranker learns from two kinds of relevance. Behavioural relevance is what people do: the results they tend to tap or download. Textual relevance is how well an app actually fits the words someone typed.

That second kind had a problem. It came from expert human judges, rating searches and apps by hand. Those labels were scarce and expensive. Far too few for a store this size.

So Apple trained a small in-house language model on the judges' answers. Then they scored how well it reproduced the judges' labels. 1.0 would be perfect.

A model ten times bigger, but not trained on those judgments: better or worse? Much worse. The bigger model scored 0.38. The small trained one scored 0.80. So training on human judgment mattered more than size.

There's also a dial in the paper. Apple can blend the two kinds of relevance in different mixes. The paper tests several mixes. It doesn't say which one shipped.

Here's the part that's easy to misread. That model doesn't rank anything when you search. It works offline. It wrote millions of relevance labels, and those labels trained the real ranker. The paper is about how those labels are made. The rest of the ranker stays private.

And it doesn't name the fields its judge reads. So anyone claiming the AI reads your screenshots is guessing.

Our take: when someone says an AI now ranks the App Store, ask them which paper says live.

Why does +0.24% matter?Watch from 4:05

So did those better labels change anything people could see? Apple tested the new ranker on worldwide traffic, against the old one. The share of search sessions that ended in at least one download rose, by 0.24%.

That's a relative lift, and Apple's researchers called it statistically significant. So it's small, and it's real. On a store this size, a small lift in that share adds up.

In how many App Store countries did the new version win? 89% of storefronts. A storefront is one country's App Store. So this wasn't a lucky result in one market.

And this is the line that matters for you. The biggest gains came in the tail. The tail is all the rare, specific searches. Few people type each one, but there are a great many of them.

On a big search like pdf scanner, the leaders already have years of downloads behind them. On a rare search there's little behaviour to go on. Our reading: the words matter more there. That's where a small, new app can actually compete.

Our take: a newer app should write for the long tail first. The big words come later.

Does retention move your rank?Watch from 5:17

But if behaviour counts, doesn't everything users do count? Like how long they keep your app? This is the most repeated claim in App Store optimization right now. Let's badge it.

In the paper, behavioural relevance is built from clicks and downloads. That's it. The words retention and uninstall don't appear anywhere in it. So retention as a ranking input gets unproven. That isn't proof Apple ignores it. Nobody has shown it.

Next, the new app boost. Some practitioners report that new apps get a temporary lift, mostly in the category charts. Apple has never said so, and no source gives a duration. Unproven.

Our rule anyway: ship your best listing on day one. You only launch once, so make it count. And don't change it in the first two weeks unless something is broken. Otherwise you can't read the launch.

Download velocity, and how often a tap turns into a download? Practitioners broadly agree those help. It fits the paper's download labels. But Apple hasn't said it, so the badge is observed.

Same for weighting. Practitioners agree the name counts most, then the subtitle, then the keyword field. Apple has never published those weights. Every chart you've seen with exact percentages is somebody's estimate.

And does Apple read the text in your screenshots for ranking? Nobody has shown it.

Our take: improve retention because it pays your bills. Just don't promise anyone it moves your rank.

Did my rank drop, or did the page change?Watch from 6:46

If your search downloads slipped this year, was it your rank, or the page around it?

Look at the page itself. In December 2025, Apple announced more ads in search results, further down the page. They went live in March 2026. Advertisers can't pick the slot. One vendor, Adapty, places the extra ad around the third result.

Now watch Coral. Say it ranks third, organically, for pdf scanner. With an ad on top and another in the middle, third place is now the fifth card down. Did Coral's rank drop?

The rank didn't move. The page did. That's a layout change, and it can look exactly like a ranking drop.

One more thing from Apple. An ad for an app that isn't relevant won't show, however much you bid. So the same words help decide whether you can rank, and whether your ad can show.

And AppTweak's monitor reported no App Store algorithm change from early August to early September.

So when downloads fall, check three things first. Did you release anything? Did your own ads change? And did total search downloads fall, or did more of them just get credited to your ads?

Our take: before you rewrite anything after a drop, check whether the page changed around you.

What decides where your app shows up?Watch from 7:59

So, back to the question we started with. What actually decides where your app shows up? Here's the working model we use. It's built on what Apple has said, but it's our model, not Apple's.

Stage one is getting into the race. Your words decide whether you're a candidate at all. Stage two is the order. Among the candidates, text fit and behaviour decide who goes first. Words decide whether you can rank. Words and behaviour on that search decide where.

Remember the search Coral couldn't enter? Search pdf scanner, and Coral is in the race. Search scan receipts, and it isn't there at all. Coral scans receipts perfectly well. But the word receipt appears nowhere in its listing. Without that word, Coral wasn't a candidate here. Downloads on other searches don't change that.

That's why the order matters. Stage two can only sort the apps that stage one let in.

So Coral ships one change. The subtitle becomes "Scan documents and receipts". Three weaker words in the keyword field make room for expense, tax and invoice. It doesn't repeat Coral or its company name: those are already searchable. And no rival app names, anywhere. Every word is something Coral truly does. A human judge would call it a fit.

After the release, where does Coral land for scan receipts? Forty-first. That's the honest outcome. Coral got into the race. It hasn't earned the order yet. Forty-first is a start. Before the change, Coral wasn't in that race at all.

Moving up is stage two's job: taps and downloads on that exact search, plus your rating. That takes time, and it takes people choosing Coral on that search. Words can't buy it.

So the model gives a small app three changes, in this order. One: make sure every search you can truly serve has its words somewhere in your title, subtitle or keywords. Two: aim at the specific searches first. That's where the gains in Apple's test were biggest. Three: win the tap and the download on those searches. Your first screenshots and your rating do that work.

And one thing not to do. Don't add words you can't serve. They waste space, and a judge wouldn't call them a fit.

Our take: write your listing for a careful human reader. Apple's paper trains its ranker on labels that copy careful human judges.

What can't this tell you?Watch from 10:18

What can't any of this tell you? Apple's list ends with "and more". And the paper covers how labels are made. It doesn't list every signal.

Our model is a model. Nobody outside Apple can see the weights. And none of this is fixed. Apple can change the ranker without telling anyone.

Coral's ranks are illustrative. They show the method, not proof that it works. Treat every observed badge as a good bet. Treat every unproven one as a test you'd have to run yourself.

Coral came in with a listing that could only enter one race. It leaves with words for the searches it can actually win.

Which keywords are worth putting in your listing? That's App Store Keyword Research 2026: The Scores Broke. Do This.

Sources

  1. Apple Developer, App Store search
  2. Apple Developer, Creating your product page
  3. App Store Connect Help, Platform version information
  4. Apple Ads, App Store (almost 65% of downloads happen directly after a search; App Store worldwide, 2022 data)
  5. Apple Ads news (additional ads in search results) (17 December 2025)
  6. Apple Ads Help, Search results ad placement
  7. Christakopoulou et al., Scaling Search Relevance: Augmenting App Store Ranking with LLM-Generated Judgments (Apple; arXiv 2602.23234, v5) (7 June 2026)
  8. Adapty, Apple Ads multiple ad slots (updated 30 Jul 2026) (4 March 2026)
  9. AppTweak, ASO news and algorithm monitor (updated 19 Sep 2026)
  10. AppTweak, App Store ranking factors (January 2026)
  11. App Radar, ASO ranking factors (April 2026)
  12. SplitMetrics, App Store ranking factors (September 2025)
  13. AppTweak, Why keyword rankings suddenly drop (updated Nov 2023)

Caveat: +0.24% is a relative lift in the share of search sessions with at least one download, from Apple's test; the paper describes how relevance labels are made, not the whole ranker, and does not say which version shipped. F1 scores are agreement with human judges' labels (best configuration per model). The two-stage model is our working model, not Apple's. Coral's ranks are illustrative.

Watch next