Answers one question per product category: which search terms should we be tracking organic rank on, and what is each one worth?
Rank trackers drift toward brand terms while the revenue sits in generic head terms nobody added. This finds those terms, prices them, and tells you which ones your tracker is missing.
Built as an Agent Skill. Runs in Claude Code, Claude web, Claude desktop and ChatGPT from the same folder.
This skill reads your live Amazon account through the TrackIQ MCP — 16 tools connecting your AI assistant to Amazon data:
Sales & Traffic · Orders · Inventory · Returns · Sponsored Products · Sponsored Brands · Sponsored Display · Amazon DSP · AMC Cloud · Keywords · Search Terms · Targeting · Search Query Performance · Organic Rank · Best Seller Rank · Buy Box History · Brand Analytics · Export
Works with Claude, ChatGPT and Cursor. Get access →
One report, three views: the cover and headline counts, a category's ranked terms with the tracking gap, then the method and limits.
One self-contained .html report.
| Section | What it does |
|---|---|
| Cover | Categories, priority terms, revenue on them, and how many are untracked |
| Headline | The account-level table: ASINs, revenue, ASP, queries seen and terms passed per category |
| How to read the action column | What Defend, Grow and Watch mean, in the report itself |
| Per category | The top 25 priority terms — weekly volume, orders, estimated monthly revenue, impression share, CTR, CVR |
| Tracking gap | How many priority terms have no rank tracking, and the revenue sitting behind them |
| Method and limits | What the numbers are, and explicitly what they are not |
Every term is labelled:
- Defend — converts and holds above-median impression share. Already won; losing rank costs money now.
- Grow — above-median demand, below-median share. The opportunity.
- Watch — small but real, on the list so a change is visible.
- The TrackIQ MCP, for
get_search_query_performance,get_product_performance,get_product_categories_performance,get_keyword_rankandlist_marketplaces - SQP ingestion running on the account. It's a separate weekly feed and is often sparse or absent on newer brands. The skill checks before promising a report.
- Nothing else. No filesystem, no shell, no internet.
Without the MCP it works from a Brand Analytics Search Query Performance export (ASIN view, one file per week) plus revenue and orders by ASIN. Everything except the rank-tracker column is reproducible that way.
/plugin marketplace add TrackIQ-HQ/amazon-seller-skills
/plugin install trackiq-amazon-category-priority-keywords@trackiq
- Download the
.zipfrom the latest release - Settings → Capabilities → Skills (code execution must be on)
- Create skill → Upload a skill, choose the
.zip - Toggle it on
Same zip. Plugins → Skills → Create → Upload from your computer.
Answers live in account.md, copied from
assets/account.example.md.
Every TrackIQ skill reads the same file, so an account already set up for
the daily or weekly send needs nothing added.
Where the report goes is asked once and stored in account.md: in-chat
(default), file, Slack, n8n or email. Anything leaving the chat
confirms with you first, and falls back to in-chat — with a note — when the
configured channel isn't available.
On onboarding, when the catalogue changes, and quarterly after that. Not weekly. The list should be stable between runs; if it isn't, something changed in the catalogue or the category rules.
This decides which keywords to track. The daily and weekly reports tell you how you're ranking on the ones already tracked.
| To change | Edit |
|---|---|
| Brand, categories, delivery | account.md — no skill edits |
| Categorisation rules, gates, labels, revenue maths | assets/method.md |
| The pull sequence and its traps | assets/pulls.md |
| The pre-send checks | assets/checks.md |
| The report shell | assets/report-template.html |
Four rules are load-bearing and worth leaving alone.
Discover which SQP weeks exist before using any of them. The feed is
intermittent and a missing week returns {"rows": []}, not an error. Assuming
four weeks when two arrived silently halves every revenue figure.
Never allocate a category's whole revenue across its queries by purchase share. It ties out to a clean total and overstates every keyword by roughly 4×, because search accounts for only 10–35% of orders.
Revenue per keyword is derived, not measured. SQP reports purchases, never revenue. The report states all three assumptions behind the estimate.
Relevance is a judgement, and it gets recorded. No metric separates a head term from a high-volume coincidence — a query can carry 107,000 searches, nine purchases, and be irrelevant. Dropped terms and the reason are written down so the next run doesn't re-litigate them.
python scripts/validate.py # must exit 0 before any commit
python scripts/build.py # writes dist/ zip + registry.jsonRead AUTHORING.md before proposing changes.
MIT. See LICENSE.