How the scores work

AsoKey has no private install data, and neither Apple nor Google publish search volume. Every number here is derived from public store endpoints, so it is an estimate built from observable signals. The breakdowns are shown for every keyword so you can judge them instead of trusting a black box.

Where the data comes from

App Store: Storefront search for the order the App Store app shows, the store listing for the name, subtitle, all-version rating count and in-app purchase flag, and iTunes Lookup for the precise star average and the last update. Apple's search hints endpoint uses the storefront of the selected country. Written reviews come from the public customer-reviews RSS feed.
Google Play: The public Play Store web endpoints for search, app detail pages, query suggestions and reviews, scoped to the selected country and language.
Ratings by country: The same public lookup used for an app's detail page, repeated across the 50 supported storefronts. A missing listing means the app is not for sale there, not a zero-star score.
Review summary: Recent written reviews from the selected country and a handful of other high-volume storefronts. An LLM (via the Vercel AI Gateway, using a cheap nano model) extracts recurring pros, cons and noteworthy comments. It is a theme read, not a complete census of every review. The summary is stored in Vercel Blob and refreshed at most once a week for the same app.
Caching: Store responses are cached in memory for one to twelve hours so repeated research is fast and the stores are not hammered. AI review summaries persist across instances in Vercel Blob for seven days.

Traffic (0-100)

Store autocomplete · 77% of measured demand: Whether the store suggests the phrase, after how few characters, and how high in the list. Stores order suggestions by real query popularity, and that list falls off steeply, so first place is worth several times fifth. Rank and prefix are multiplied: a phrase that only appears once the whole query has been typed is a weak signal, even if it is then first. A phrase the store never suggests keeps a small floor, because long-tail queries are still typed.
Audience of named apps · 23% of measured demand: Median rating count, on a log scale, of apps that put this exact phrase in their title. One app is enough: a single brand that owns the name is real demand. Apps that merely rank for the query, including a giant that never targeted the words, contribute nothing here, so their install base is not mistaken for search traffic. When the store returns results and nobody uses the phrase, this component is zero. It is dropped only when no results came back at all, and the other weights are then rescaled.
Query breadth · a scale, not a bonus: After the two measured signals are combined, phrase length scales that result from 82% for a long query to 100% for a single word. A short word the store never suggests does not gain traffic from being short.

Difficulty (0-100)

How the number is chosen: The headline is whichever is higher: the top-5 rating barrier, or the weighted mix of all five signals. Weak title matches cannot talk a huge rating wall down into a moderate score.
Top-5 rank barrier · 36%: The rating volume a new app has to beat to take a top-5 slot. The weakest of those five counts most, because that is the listing you displace; the median stops four giants and one empty result from looking wide open. Fewer than five results means the missing slots are unoccupied. Counts are log scaled from about 100 ratings (no real barrier) to about five million (a category-defining incumbent).
Title ownership · 28%: How completely the top 5 use the phrase in the app name, weighted towards the first positions. An exact title counts fully, the same words in any order a bit over half, and a partial overlap a fifth. A subtitle, or a Play short description, counts at less than half of the same match in the name. Matches are whole words, so a fragment hiding inside a longer word does not count. Empty slots count as unowned, so a single matching app on an otherwise empty page is not “fully taken”.
Active defence · 14%: Of the top-5 apps that target the phrase in the title, the share updated in the last four months. An abandoned exact-match title is easier to pass than one still shipping.
Incumbent tenure · 12%: How long those title-targeting apps have been listed, up to four years. A phrase owned by apps launched this year is easier than the same ratings on a four-year incumbent. If nobody targets the title, tenure is zero rather than unknown.
Rating quality · 10%: Average star rating of established apps in the top 5.

Chance

Geometric mean: chance = √(traffic × (100 − difficulty)) — a keyword has to be both reachable and worth reaching. Anything that scores well on one axis alone stays mid-table. Traffic and difficulty no longer share the same rating-volume figure, so a hard head term and an open suggestion can move independently. If the store returns no apps, that is not treated as an open market: chance stays a small fraction of traffic, so a word nobody ranks for cannot look like an opportunity.

Monthly revenue

A range, not a store report: Apple and Google do not publish what an app makes. AsoKey estimates net revenue for a month in US dollars, and shows the range rather than a single figure. It can be off by several times. The pieces below are the whole model.
Lifetime audience: On Google Play, the install count (the public bucket, or the closer count inside it when Play publishes one). That number is worldwide. On the App Store, ratings stand in for installs: 50–200 installs per rating, which is centered on the same one-rating-per-hundred-installs ratio used elsewhere. A single lookup is that storefront only. The competitor teardown adds the rating counts from the supported markets once they load.
Which installs landed this month: Lifetime installs divided by the app's age in months, then multiplied by how recently it shipped. The share is capped at the app's whole lifetime. An app released in the last 30 days keeps 85–100% of its installs in the current month. When the release date is known, that monthly share is widened to 60–160%. When it is missing, age is assumed at 36 months and the band widens to 35–280%.
Updated in the last 45 days: Monthly installs are about 2.4× a flat lifetime-divided-by-age split.
Updated in the last 4 months: Monthly installs are about 1.6× a flat lifetime-divided-by-age split.
Updated 4–12 months ago: Monthly installs are about 0.8× a flat lifetime-divided-by-age split.
No update for over a year: Monthly installs are about 0.25× a flat lifetime-divided-by-age split.
What one install is worth: Net dollars to the developer. A paid app uses its store price after the standard 30% commission (the developer keeps 70%), converted to USD with a fixed rate table. Everything else uses the category blend below. Play listings that declare in-app purchases sit toward the top of the blend; ads-only listings sit toward the bottom; a free app that declares neither is narrowed further. The App Store listing says whether an app sells in-app purchases, so iOS uses that the same way. It does not say whether the app runs ads, so an iOS app without in-app purchases keeps the full blend.
Games: $0.15–$1.80 net per install.
Finance: $0.40–$3.50 net per install.
Medical: $0.30–$2.50 net per install.
Productivity and tools: $0.25–$2.20 net per install.
Social: $0.04–$0.40 net per install.
Education and news: $0.08–$0.90 net per install.
Entertainment: $0.06–$0.70 net per install.
Travel: $0.08–$0.80 net per install.
Lifestyle: $0.10–$1.20 net per install.
Everything else: $0.05–$0.60 net per install.

App keyword discovery

Listing metadata: The same extraction used in competitor teardowns: phrases from the app name, short description and full description, weighted by field and repetition.
Store autocomplete: Those seeds are expanded through the store's own suggestion endpoint so the shortlist includes phrases people actually type, not only what the listing contains.
Live rank check: Up to 25 candidates are searched in the selected country. Only phrases where the app appears in the scanned results are returned, ordered by position.

Known limits

Monthly revenue is an estimate: The range is built from install counts or ratings, the app's age and last update, and a category value per install. It is not a payout report. iOS has no public install count, Play buckets are coarse, and a category blend is not that app's price list.
Search results are a storefront, not a signed-in phone: Rank and the apps behind a score come from storefront search, the same list the App Store app loads for a logged-out shopper in that country. A signed-in device can still differ, because Apple personalises and A/B tests results.
iOS keyword field: The subtitle is on the public listing. Apple still does not expose the private 100-character keyword field, so competitor teardowns cannot show it.
Play rating counts: Google Play search results omit rating counts, so they are pulled from each app's detail page. Where only an install bucket exists, roughly one rating per hundred installs is assumed.
Discovery is a shortlist: Automatic keyword discovery probes the strongest candidates from the listing and autocomplete, not every possible query in the store. Add extras on the Keywords screen when you want a specific phrase checked.
Reviews are a sample: Apple's public RSS feed caps written reviews at a few hundred recent comments per country. Play returns a language-scoped page of recent reviews, not every rating. The AI summary therefore reflects recent, vocal users more than the full history.