Finding keywords

The four ways the Discover tab brings terms in, what each costs, and why the list clears itself.

Four sources

• Paste — terms you already have, one per line. Costs nothing to collect.
• Harvest — asks Play's autocomplete what people type after your seed, expanding letter by letter in the app's own alphabet. Up to 40 queries per run.
• AI ideas — sends your listing to the provider whose key you added and asks for search terms. Needs your own key.
• Reviews — reads the 150 newest reviews of your app or a competitor and extracts the words users repeat. Review text is never stored.

Collecting is where the four differ. Harvest and Reviews spend Play requests to collect; Paste spends none, and AI ideas go to your own provider rather than to Play.

Measuring is where they are the same. Every term that reaches the pool is queued for a rank check and, when it has no suggestion evidence yet, an autocomplete probe — so Paste and AI ideas do spend Play requests too, from the same shared ceiling as everything else. At most 100 keywords enter measurement per action; anything beyond that is reported back to you rather than queued.

Reviews are filtered against your listing

Reviews are sentences, so a naive extractor returns verbs and adjectives rather than search terms. A single word is accepted only if it also appears in your own listing; terms that start or end with a rating word are dropped; reviews below three stars are excluded, because they produce a complaint vocabulary rather than a search vocabulary.

If you have an AI key, one more filter runs: the model is asked which candidates look like real Play searches. It can only remove, never invent, and if it removes everything the unfiltered result is used instead — an empty scan is worse than a noisy one.

The discovery pool is a workbench, not an archive

A new scan clears the previous discoveries for that app, and whatever is left expires after an hour. The pool is where you decide; the tracking list is where things live.

Nothing measured is lost when the pool clears. Scores and rankings are stored globally, so a term you discover again comes back with its numbers already filled in.

Dismissing a term is permanent: harvest, AI and review scans skip it from then on. Pasting it by hand still works, because that is an explicit request.

Everything discovered gets measured

Each new term is queued for a rank check and, if there is no suggestion evidence for it yet, an autocomplete probe. The strip at the top of Discover shows the two stages: reading the terms, then measuring popularity and difficulty.

A row counts as finished only when both popularity and difficulty are filled. Polling runs for about sixteen minutes; after that the strip says how many are still measuring and the rest complete on their own.

→ The keyword pool