Sourcing across five signals
Apple Search Ads impression reports (the truest volume signal Apple releases). Competitor teardowns (AppTweak, Sensor Tower, Mobile Action). Review-mining for user vocabulary. Paid-search harvest from web SEM. Assistant-prompt panels for AI-Discovery vocabulary. Never trust one source.
Scoring: difficulty × volume × intent
Every candidate keyword gets three scores: difficulty (competitive ranking density), volume (impression-report signal), and intent (buyer proximity, 'best budget app' beats 'money'). Multiply for a composite priority score. Cut anything under threshold.
Mapping to fields
Apple: title (30), subtitle (30), keyword field (100). Play: title (50), short description (80), full description (4000). Every keyword goes in exactly one field, no duplication. Higher-priority terms live in higher-weight fields.
The refresh cadence
Track ranks regularly, refresh the keyword set on a monthly cadence, revisit semantic surfaces quarterly and review positioning annually.
Long-tail vs head
Head terms can build visibility while long-tail terms often carry clearer intent. Set the mix from observed category demand and revise it as the listing earns signal.
Cross-market keyword hygiene
Never translate keywords. Harvest local buyer language: 'coche derrape' in Spanish, 'katta racer' in Hindi-English, 'sfeeratun' in Egyptian Arabic. A translated keyword field is a dead keyword field.
Common failures
Discovery-once, refresh-never. Trusting a single tool. Optimising for tool difficulty scores rather than real category signal. Ignoring intent in favour of volume. Duplicating terms across title and subtitle.