Amazon Listing & Ads Skill Suite: From SKU Evidence to Paused Ad Drafts

#amazon #ecommerce #ai #automation

Canonical bilingual article: https://www.jiajingqiu.com/amazon-listing-ads-skill-suite/

Amazon listing and advertising work usually breaks in one of two places: content is drafted before the SKU evidence is complete, or an advertising “plan” quietly becomes a live account change before anyone has reviewed it.

I rebuilt four working packages into one modular skill suite that keeps those boundaries explicit. It takes a product from evidence collection to listing copy, paused campaign drafts, and report analysis—without inventing product claims and without writing to an Amazon Ads account.

Six focused skills

Amazon Listing Collector — captures SKU facts, source provenance, competitor structure, and missing-evidence flags.
Amazon Review Analysis — separates verified customer language, recurring objections, use cases, and unresolved risks.
Amazon Keyword Library Builder — consolidates and classifies search language without prematurely assigning ad match types.
Amazon Listing Generator — turns evidence into compliant title, bullets, description, and backend-search-term drafts.
Amazon FAD Initialization — owns campaign structure, match type, bids, budgets, and Amazon Bulk draft generation.
Amazon Ad Report Analyzer — reads one supported Sponsored Products or Sponsored Brands CSV, recomputes performance metrics, and produces reviewable action candidates.

The root skill routes work to the correct module, so collection, content generation, campaign planning, and performance analysis stay separate.

The operating sequence

The workflow is intentionally staged:

SKU evidence → review evidence → keyword library → listing draft → paused ad draft → report analysis

Each stage produces a structured artifact that the next stage can inspect. If a product claim has no source, it remains a gap. If a report type is ambiguous, analysis stops instead of guessing. If a recommendation could change spend, it stays pending human review.

Safety contract

The suite is designed for production preparation, not unattended account mutation:

Every stateful Amazon Bulk row is generated as paused.
Positive keyword match types are assigned only in the FAD layer; upstream research skills emit neutral ad candidates.
Campaign structure, match type, bid, and budget decisions have one clear owner.
The analyzer recalculates CTR, CPC, CVR, ACOS, and ROAS from raw fields rather than trusting spreadsheet formulas.
Zero-sales rows report ACOS as NA, not an artificial zero.
Analysis outputs are marked pending_human_review.
There are no Amazon credentials, network writes, account mutations, or automatic Bulk uploads in the public package.

Policy-sensitive rules are kept in dated references. For example, Amazon announced US listing-title and item-highlight changes taking effect on July 27, 2026; the suite treats that as a dated marketplace constraint rather than a timeless copywriting rule.

Worked food-storage example

A checked demonstration for a food-storage-container SKU produces a 51-row paused Bulk draft across six planned campaigns. Its configuration uses a USD 30 daily ceiling, with USD 27 allocated and USD 3 held in reserve. The readiness gate is 66/100, so the output remains a draft for review.

Those numbers describe the planning artifact. They are not GMV, CTR, conversion, or performance claims.

What is public

The bilingual methodology, module boundaries, safety rules, and worked-example interpretation are public on my portfolio. The implementation page and downloadable ZIP packages are access-controlled, while the GitHub source repository remains private.

Read the full Chinese/English case study:

https://www.jiajingqiu.com/amazon-listing-ads-skill-suite/