App Store Optimisation in the AI-Search Era

App store optimisation has a reputation as keyword stuffing with a respectable name. That reputation was earned, and it is now actively counterproductive — not because the stores changed dramatically, but because a growing share of app discovery no longer starts in a store at all.

The mechanics of store search still reward roughly what they always did. What changed sits upstream, and most teams are still optimising only the final step.

What still works, unchanged

The fundamentals are stable, and they are where most of the available gain still is.

Your title and subtitle are the highest-weighted text you control. Not a slogan: what the app is, in the words someone would search. "Zen" tells the store nothing. "Zen: Sleep Sounds & Timer" is findable.

Keyword fields are for terms not already in your title. Repetition is wasted space, and the stores handle plurals and word order themselves.

The first screenshot does most of the conversion work. Most people never swipe. It should show the core value, legibly, at thumbnail size, which is smaller than whatever you are designing on.

Ratings gate everything. Below a certain point no amount of optimisation compensates, because ranking and conversion both depend on it. Ask for reviews after a good moment in the app, never on launch.

Retention feeds ranking. Stores promote apps people keep. Which means the highest-use ASO work is often not marketing at all: it is fixing the first-run experience so people come back.

If your ratings are poor and your retention is weak, ASO is not your problem. The app is, and optimisation will only bring more people to the same experience.

What changed

The shift is that a meaningful share of "which app should I use for X" now gets answered by an assistant, a search engine's generated summary, or a community thread. Before anyone opens a store.

That matters because those answers are not built from your store listing. They are built from what exists about your app elsewhere: your website, reviews and comparisons, forum discussions, documentation, coverage. A perfect listing is invisible to a recommendation process that never reads listings.

Which produces an uncomfortable conclusion for teams who have treated ASO as a store-only discipline: the store listing is now the last step of discovery, and increasingly not the first.

Being findable by systems that summarise

Practical implications, none of them exotic:

Have a real web presence for the app. A page that states plainly what it does, who it is for, what it costs, and what platforms it runs on. Assistants and search engines can read a web page; they cannot read your App Store listing nearly as readily.

Answer the comparison question yourself. People ask "best app for X" and "X versus Y". If you have never written honestly about where your app fits and where it does not, that question is being answered entirely by other people.

Write in plain, specific language. Systems that summarise reward clear factual statements over marketing register. "Tracks daily habits and generates a weekly review" is more useful to a summariser, and to a human, than "reimagine your potential".

Keep facts consistent everywhere. Pricing, platform support and feature lists that disagree between your site, your listing and your documentation produce hedged, low-confidence recommendations.

Structure the page properly. Clear headings, straightforward answers to the obvious questions. The same discipline that makes a page easy for a person to skim makes it easy for a system to extract.

This is where app marketing and ordinary web development have converged. The app's website is no longer a brochure; it is the source most recommendation systems read.

Two settings that are quietly strategic

Your category. It determines which chart you appear in and who you are compared against. The instinct is to pick the largest category, which is usually wrong — visibility in a smaller, accurately-matched category beats invisibility in a crowded one, and the users who find you there arrived with the right intent. Where two categories fit, the deciding question is which one your ideal user browses, not which one has more traffic.

Your icon. It is the single most-seen asset you own and the one most often designed at the wrong size. It appears at thumbnail scale in a list of competitors, so it needs one clear shape and enough contrast to survive both light and dark backgrounds. Words in an icon are almost never legible at the size it is viewed. If you test one thing on your listing, test this.

Both are easy to change and rarely revisited after launch, which makes them a cheap source of improvement for a product that has been in the stores for a while.

Reviews are now training data for recommendations

Reviews have always affected ranking and conversion. They now also feed what gets said about your app in a summary.

The practical consequence is that what reviews say matters as much as their average. Fifty reviews all praising the same specific thing establish what your app is known for far more effectively than five hundred saying "great app". Prompting for reviews right after the moment your app delivers value tends to produce specific ones, which is both more honest and more useful.

Responding to negative reviews matters for the same reason. An unanswered complaint is the only account of that issue; a response describing the fix changes what the record says.

Testing a listing without guessing

Most listing changes are argued about rather than measured, which is odd given that both stores provide a way to test them properly. If you can run a controlled experiment on your store page, the arguments become unnecessary.

A few rules make the results usable:

• One variable at a time. Changing the icon, the first screenshot and the subtitle together tells you the combination performed differently and nothing about which part did the work.
• Run it long enough to cover a full weekly cycle. Weekday and weekend traffic behave differently in most categories, and a three-day test reliably produces a confident wrong answer.
• Test the things with the most use first: icon, first screenshot, title. These are what someone sees in a search result before deciding to tap.
• Judge on installs, not impressions. A change that lifts impressions while lowering conversion has made your listing more visible and less persuasive, which is usually a step backwards.

Where a controlled test is not available, smaller markets, or too little traffic for significance: change one thing, hold it for a few weeks, and compare against the equivalent prior period rather than the week before.

Paid and organic are not separate programmes

Teams tend to run these as two disconnected efforts, which wastes the most useful property they have: paid traffic tells you quickly what organic would have taken months to reveal.

Paid campaigns give you fast, statistically meaningful data on which messages convert and which keywords attract people who install. That is directly transferable to your listing text, your screenshots and your keyword choices. Running ads while never feeding the learnings into the organic listing is paying for research and discarding it.

The relationship runs the other way too. Installs driven by paid activity contribute to the ranking signals that improve organic placement, but only if those users behave well after installing. Buying installs from an audience that churns immediately can leave you worse off than before, because retention feeds ranking and you have just diluted it.

Which brings the argument back to the same place: acquisition of any kind multiplies against product quality. If retention is weak, both channels underperform, and the highest-return work is not in either of them.

What has not changed at all

Localisation remains among the highest-return work available. Translating your title, keywords and screenshots into the languages your users speak opens search terms nobody in your market is competing for. Machine translation for keywords, reviewed by a speaker. The review step is what stops you ranking for something unintentionally funny.

Seasonality still works. Screenshots and copy adjusted around the moments your category spikes still outperform a listing left untouched for two years.

And measurement still tells you what happened. Impressions, product page views, conversion, and retention by cohort, without those, ASO is decoration. Change one variable at a time and give it long enough to read.

Where to start

For most teams, in this order:

• Fix ratings and first-run retention. Everything else multiplies against these.
• Rewrite the title and subtitle in the words people search, not the words in your pitch deck.
• Redesign the first screenshot for legibility at thumbnail size.
• Build the app's web page properly, in plain factual language — this is the part most teams skip and where the newest opportunity sits.
• Localise into your top non-English markets.
• Then iterate keywords, one change at a time, measured.

Note that only steps 2 and 6 are ASO as it is usually described. The rest is product quality, design and web presence, which is the real point. Apps like Zoolingo reached millions of downloads on the strength of retention and word of mouth as much as listing text, and our app store optimisation work starts from the product for exactly that reason.