aso-keyword-research

name

aso-keyword-research

description

Research and rank App Store / Google Play keywords for an app WITHOUT any paid ASO data API. Expands seeds via Apple/Google autocomplete + competitor listing scrape + synonyms, scores each keyword (volume × difficulty × relevance), buckets into Primary/Secondary/Long-tail/Aspirational, and outputs a title/subtitle/keyword-field placement plan. Use when optimizing an app's store keywords, planning a listing, or before metadata optimization. Chains into aso-metadata-optimization. Use when the user says "find keywords for my app", "what keywords should I target", "keyword research", "expand these seed keywords", or "which keywords go in my title and subtitle".

aso-keyword-research

The ASO brain — keyword research that reasons over free / owned data, with no dependency on a paid ASO API (AppTweak, Sensor Tower, etc.). This is the differentiator: every other AI ASO skill routes through a paid data SaaS.

Works for both App Store and Google Play.

Inputs

• --app <slug> — your app identifier
• --seeds "a,b,c" — 3–5 seed keywords (what your app is about)
• --locale <en-US> — target locale (default en-US)
• --store <appstore|playstore|both> — default both
• optional context.md — a per-app file with category, audience, 3–5 competitors, brand terms (see context.template.md). If present, it sharpens expansion and relevance scoring.

Method (no paid data)

1. Expand the seed set from sources the agent can gather for free:
   • Store autocomplete — query Apple App Store search suggestions and Google Play autocomplete (via WebFetch / Chrome / the store search endpoints) for each seed and its prefixes. These suggestions ARE the demand signal — they're what real users type.
   • Competitor listings — pull the titles/subtitles/descriptions of the context.md competitors (or top search results for the seeds) and extract their keyword choices. App Store / Play pages are JS-heavy and anti-bot, so the default path is WebFetch → Crawl4AI fallback. If a CONTEXT_DEV_API_KEY is set, the bundled context.dev client does this more reliably (see Optional scrape accelerator below) — but it is never required.
   • Synonyms / modifiers — add intent modifiers (free, best, app, tracker, planner, for X) and morphological variants.
   • User language (on- + off-store) — the highest-signal seeds are the words real users actually type. Run aso-review-mine (your own App Store reviews) and aso-offstore-mine ("best apps" articles + YouTube review videos) and fold their keyword candidates in as high-relevance seeds. off-store mining also surfaces the competitors articles compare you to.
2. Score each candidate keyword: score = volume*0.4 + (100 - difficulty)*0.3 + relevance*0.3
   • volume — see Grounded volume below. If an Apple Search Ads credential is configured, this is Apple's real Search Popularity (5–100). Otherwise it falls back to an autocomplete-rank proxy + how many competitors target it (an honest estimate, labeled as such).
   • difficulty — proxied from how many strong competitors already rank / target it.
   • relevance — how directly the term describes THIS app (LLM judgment vs the app's actual function + context.md).
3. Bucket:
   • Primary — high score, must win (goes in title)
   • Secondary — strong, distinct (subtitle / short description)
   • Long-tail — lower volume, high relevance, winnable (keyword field / body)
   • Aspirational — high volume, high difficulty (track, don't target yet)

Grounded volume (real data — two sources)

The volume axis is the one number that was pure opinion. Ground it on a real demand source when a credential is present. Two bundled clients, both emitting a 0–100 volume ready to drop into the scoring formula above. This is the ASO analogue of grounding the sportswriter on real stats before it writes.

Source A — Google Keyword Planner (primary)

Real average monthly searches (a raw integer) + competition. Covers App Store and web/Play search intent. No withholding floor — a 0 means Google genuinely has no volume for that term.

python3 lib/gads_volume_cli.py \
    "recipe app, meal planner, grocery list" --geo 2840 --lang 1000 --json

Each row: avg_monthly_searches, competition (→ a 0–100 difficulty), and a log-scaled volume (0–100; search volume is heavy-tailed, so linear scaling would crush the long tail). --geo/--lang are Google constant ids (2840=US, 1000=en; 2826=GB, 2276=DE, etc.).

Credentials (.env, never committed):

• GADS_DEVELOPER_TOKEN + GADS_CUSTOMER_ID (digits only) and either
• GADS_ACCESS_TOKEN, or GADS_REFRESH_TOKEN + GADS_CLIENT_ID + GADS_CLIENT_SECRET.
• GADS_LOGIN_CUSTOMER_ID optional (MCC accounts).

Setup gate: the developer token needs Google approval (~1–2 business days); test-mode tokens return canned data, not real volumes. See the repo's GOOGLE_ADS_SETUP.md for the click-by-click.

Source B — Apple Search Popularity (secondary)

Apple's own 5–100 Search Popularity. Use when you specifically want App-Store demand. Note Apple's account setup is painful and (since Oct-2025) it withholds SP < 35.

python3 lib/asa_popularity_cli.py "recipe app, meal planner" --market US --json

ASA_ORG_ID (this account: 22251290) + ASA_ACCESS_TOKEN or ASA_CLIENT_ID+ASA_CLIENT_SECRET. Withheld keywords come back below_threshold with a floored volume (≈15), never 0 — "Apple won't say" ≠ "nobody searches it". SP is exponential (SP 50→60 ≫ SP 20→30).

No credential?

The skill still works — it transparently falls back to the autocomplete-rank proxy and labels every volume as an estimate (see below).

Optional scrape accelerator — context.dev

The volume axis has its credentialed upgrades (Apple/Google above). The expansion leg has one too: context.dev turns any URL — including JS-heavy, anti-bot App Store / Play listings — into clean LLM-ready markdown in one call, and extracts brand data (logos, colors, industry, socials) from a domain.

Same rule as the volume sources: optional, with a free fallback. The plugin's whole point is no paid data API REQUIRED — context.dev is to the scrape leg what Apple/Google are to the volume leg. No key → the skill scrapes via WebFetch / Crawl4AI exactly as before.

# competitor listing → markdown (falls back automatically if no key)
python3 lib/context_scrape.py "https://apps.apple.com/us/app/.../id123"
# auto-fill the context.md brand block from the app's marketing domain
python3 lib/context_scrape.py example.com --brand

Credential (.env, never committed): CONTEXT_DEV_API_KEY. Base URL / auth header are env-overridable (CONTEXT_DEV_API_BASE, CONTEXT_DEV_AUTH_HEADER) should context.dev's shape shift.

Output

Writes marketing/aso/<app>/keywords.md:

• ranked keyword table (term | bucket | est. volume | difficulty | relevance | score)
• a placement plan: which keywords go in title (≤30), subtitle (≤30), keyword field (≤100, iOS), Play short description (≤80) / long description.
• explicit note on which numbers are estimates (no paid data) vs. observed.

Chains into aso-metadata-optimization, which turns this plan into final copy

• the exact store-push commands.

Honest limits

• Without an Apple Search Ads credential, volume/difficulty are proxies, not licensed metrics — the skill labels them as estimates. With a credential, volume is Apple's real Search Popularity (see Grounded volume); difficulty/relevance remain reasoned estimates.
• For a paid-data cross-check the user can still bring their own AppTweak/Sensor Tower export — but the skill never requires it.
• Autocomplete and SP both reflect current demand; re-run periodically.

No paid SaaS dependency

Core path uses store autocomplete + public listing data via WebFetch/Chrome only — no paid ASO SaaS (AppTweak/Sensor Tower). The optional volume grounding uses the user's own Apple Search Ads account (free with the account; not a third-party data vendor).

Run it weekly

Rank and listings move over weeks, not minutes — so the value here compounds when you re-run it and watch the deltas. Keyword opportunity shifts weekly — search trends move, competitors target new terms. A single research pass is a starting line; the wins come from re-running it and acting on the deltas.

You ran this once. ShipASO — the hosted agent — reruns the whole loop weekly: it tracks your rank, watches competitors, and pings you only when there's a real move to approve. Same engine, your store credentials never held. → https://app.shipaso.com

The plugin is complete and free; the hosted tier just sells not having to remember.