Keyword Research

Keyword research → prioritized keyword map

Discovery is cheap; focus is the product. This skill ends with a per-ASIN keyword map routed into the other skills — not a 1,500-row dump.

Inputs

Your ASIN(s), 2–5 competitor ASINs (from get_listing_details of your top keywords' ranking pages, or user-provided), and optionally seed phrases.

Phase 1 — Discover (pick per goal; watch the quota)

• Reverse-search a competitor: get_keywords_by_asin (Cerebro) — every phrase

the ASIN is visible for, with volume, rank, CPC, CPR, title density. ⚠ HUGE output (can be 1,000+ rows). Use for 1–2 ASINs max per session and summarize immediately; never paste raw output back to the user.

• Gap vs competitors: `gettopkeywords(main_asin=yours,

competitor_asins=[theirs])` — multi mode returns shared keywords + which ones competitors rank top-40 for that you don't. This is the highest-signal call per token; prefer it over multiple Cerebro pulls.

• Seed expansion: get_keywords_by_keyword("stun gun") — related phrases

from a seed when there's no good competitor ASIN yet (new product research).

• Fresh angles: get_keywords_new_suggestion for terms recently gaining

volume around an ASIN.

• Score a finalized list: analyze_keywords([...]) — batch metrics for

candidate lists from any source (brainstorm, PPC search terms, etc.).

Phase 2 — Filter & prioritize

Score each candidate on:

1. Volume — floor of ~300/mo for copy targets; no floor for backend terms.
2. CPR — units needed in 8 days for page-1 top half. **CPR ≤ 15 = cheap

ranking push**; flag these separately, they're the actionable list.

1. Relevance — does the phrase describe THIS product? (Ranking for a

near-miss term brings clicks that don't convert and hurts CVR.)

1. Trend — 30-day search-volume trend; a +30%/mo term at 500 volume beats a

flat 800.

1. Competition texture — exact title-match count (low = title opportunity),

CPC (proxy for commercial intensity).

Phase 3 — Output: the keyword map (per ASIN)

• Primary (2–4): highest volume × relevance → title + first bullets.
• Secondary (5–10): remaining bullets, description/A+.
• Backend (~250 bytes): everything indexable that doesn't fit copy — no

brand names, no commas needed, no repeats of title words.

• Push list: CPR ≤ 15 terms with current rank 5–20 → hand to ads/promo.
• Tracker list: primaries + push list → for the user to add to Helium 10

Keyword Tracker (UI-only; no API write) so rank history starts accruing.

Route: copy targets → listing-guard (draft → approval → PATCH), measurement → impact-tracker baseline, coverage re-check → next store-audit.

Gotchas (learned in production)

• Variation siblings share keyword data — research the family head, not each child.
• time_period=YYYY-MM on Cerebro gives one historical month per call (24-month

lookback) — useful for seasonality checks before Q4 bets.

• A keyword you rank #6–9 on is worth more than a new one at #80: check current

rank before chasing volume (gettopkeywords single mode shows your ranks).

• Quota: a focused research session ≈ 5–10 calls of the 1,000/period. The trap

is Cerebro-ing every competitor "to be thorough" — the gap call usually answers the question in one.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: EbrahimElbagory
• Source: EbrahimElbagory/claude-fba
• License: MIT

Install and usage instructions live in the source repository linked above.