The keyword pool

The Keywords page: three tabs, the filters that matter, row actions and bulk work.

Three tabs

• Tracked — what you measure every night. This is your list.
• Measured — keywords already measured for this store and language that overlap your listing but are not in your pool. Adding one costs no Play request; the measurement already exists. Only terms Play itself suggests appear here: a keyword someone typed by hand is measured, but it never shows up in another workspace's list.
• Discover — where new terms come from.

If a list cannot be loaded, it says so and offers Retry — it never passes for an empty pool. With filters on, an empty result reads "Nothing matches these filters." — so an empty screen always means one thing.

Sorting the Measured tab

Its four columns — keyword, popularity, difficulty, relevance — are sortable; click a heading once for the highest values, again to reverse it. Keyword sorts by the rules of the app's own language, so Turkish puts ç between c and d rather than after z.

Sorting happens on the server, before the list is cut to fifty rows. That matters: the tab usually has more candidates than it shows, and sorting only what is on screen would give you the most relevant of the fifty most popular rather than the most relevant of all. A keyword whose difficulty could not be measured stays at the end in both directions — unknown is not the same as easy.

Rising terms

The Measured tab has a Rising view: keywords you don't track that overlap your listing and whose popularity estimate rose by at least 8 points over the last week, are at least 20 today, and were already up on the previous measurement. Rising rows carry a "↑ +N" badge and can be tracked from the same list.

If there is less than a week of daily measurements, the view says so instead of showing an empty result; a rise means Play now suggests the term more readily, not more installs.

Two bands worth reviewing

The rank filter has two entries that are not obvious. Almost first page is positions 11 to 20: the highest-return work in ASO is usually moving 14 to 9, not 40 to 1.

Dead weight is measured, never ranked, and low popularity — all three at once. A keyword that has never been measured cannot be judged, and one that does not rank but is popular is still a target. Keywords in this band are measured once a week rather than every night.

Near-duplicates

Above the filters, collapsed by default, a card groups tracked keywords that reduce to the same root — the inflected forms of one term sitting in your pool as separate rows, each costing a measurement every night.

It suggests which one to keep: by popularity when every keyword in the group has a measured popularity, otherwise by the best rank, and it says which of the two it used. When neither is known it suggests nothing rather than guessing.

The grouping is a suggestion, never an action. Tick the rows you mean to drop — shift-click takes a range — or let Apply suggestion tick the forms each group would drop for you. Either way you can change the selection freely before anything happens, and Remove selected sends it as one request. When no group has enough data to suggest a keep, that button says so instead of doing something arbitrary.

The confirmation names every keyword it is about to remove, one by one, rather than counting them. Removal cannot be undone, and a dialog that said "3 keywords" would hide which of two near-identical forms was the one leaving. The card itself has no per-row button; the ✕ on a Keywords row and Remove N from pool use the same kind of confirmation.

What goes is the keyword's place in this app's pool — which frees a plan slot and one Play measurement a night. What stays is the keyword itself and its rank history in the shared store, so re-adding it later brings the numbers back.

Root rules sometimes cut too deep — English plural stripping groups "news" with "new" — which is exactly why nothing is removed for you.

If the card cannot load, it says so instead of drawing nothing. "No near-duplicates" and "we could not ask" are different answers, and telling them apart is the whole point of the card.

The card appears only for languages that have a root rule: today English, French, German, Polish, Russian, Spanish, Turkish and Ukrainian. In any other language it is not drawn at all: an empty card would read as "no duplicates", when the honest answer is "we cannot tell".

Above the table, when your app's language is not fully covered, one line says what is missing in it — suffix stripping, the filler-word list, or the review filler list — and which surfaces that closes. Nothing is written when the language is complete: an "everything is fine" line on every app would stop being read. No list is invented for a language that lacks one; a made-up filler list would drop real search terms, which is worse than having no filter at all.

How much of your plan is left

The Tracked tab carries a small counter: keywords used against your limit, apps used against yours. The keyword limit is a workspace total across every app rather than a per-app allowance, because that is what actually costs — every tracked keyword is re-measured, most of them every night.

It is computed from the same queries the server runs when it accepts or refuses an addition, so the counter never promises room that the server then refuses. The add-app wizard shows the same numbers, and its Select all picks only as many keywords as will fit.

Row actions

• ★ favourite — a one-click grouping; alerts can be narrowed to favourites only.
• The score badge opens the breakdown: every signal, its weight and what it contributed.
• ⚑ report — tell us a score looks wrong. Scores are estimates and a long-standing error once went unnoticed for months because there was no way to say so.
• ↻ re-measure — queues a fresh rank check and probe for that one keyword, with a ten-minute cooldown. The result always lands in the inbox, even when the move is below your alert threshold, for example #13 (was #12).
• ✕ remove — takes it out of this app's pool, after a confirmation that names the keyword.

Working in bulk

Tick rows to act on many at once; shift-click selects a range. Selection follows what is on screen, so a filter or a page change never acts on rows you cannot see.

Bulk re-measure is capped at 25 keywords, because every one of them is a Play request. Keywords still in their cooldown are skipped and counted rather than refused.

Remove N from pool lists every keyword it is about to remove, one by one, before anything happens. Removal cannot be undone, and the button stays locked while the request runs so a second click cannot send it twice.

Export CSV downloads exactly the rows shown, with the filters and sorting you applied — every page of them, not just the visible one.

Where your keywords came from

The source scorecard above the table shows, per discovery source, how many keywords it brought and how many of them actually rank. The ratio alone is misleading: competitor-sourced keywords score lowest but put the most terms into the top ten, because they are the hard ones. First-page and threshold counts are therefore shown separately.

→ Reading the scores