Amazon PPC 搜索词审计

概述

Amazon PPC Search Term Audit

Turn report rows into traceable review candidates without hiding weak evidence or automating campaign changes.

Workflow

1. Confirm ad product, report type, marketplace, currency, report grain, date range, attribution window and maturity, advertised products, campaign goal, target ACOS, break-even ACOS, minimum evidence, and protected brand or strategic terms.
2. Read [references/report-schema-and-quality.md](references/report-schema-and-quality.md) before mapping columns or combining files. Keep marketplaces, currencies, attribution windows, and materially different ASIN economics separate.
3. Read [references/action-framework.md](references/action-framework.md) and record all thresholds used. Do not silently reuse a target ACOS or break-even ACOS from another product or marketplace.
4. Run scripts/analyze_search_terms.py against the CSV when the input is tabular. Use decimal ACOS values, such as 0.30 for 30%, and set locale separators or column overrides when required.
5. Review mapped columns, rejected or anomalous rows, totals, duplicate terms across campaign contexts, and attribution completeness before interpreting classifications.
6. Separate keyword-like queries from product-identifier-like terms. Preserve campaign, ad group, target, match type, advertised ASIN, metrics, action, evidence status, and reason whenever those columns exist.
7. Treat harvest, negative, product-target, and bid-down outputs as candidates. Check relevance, brand defense, campaign intent, ASIN mix, match-type leakage, placement, stock, price, listing changes, seasonality, and cannibalization before action.
8. Prioritize by commercial impact and evidence quality, then state a next-review condition for observations and protected or attribution-incomplete rows.

Deliverable

Provide:

• report scope, column mapping, thresholds, and data-quality limitations;
• KPI summary with spend, sales, orders, ACOS, ROAS, CPC, CTR, CVR, and CPA when inputs permit;
• action and evidence-status counts;
• a prioritized candidate table retaining campaign context and reasons;
• wasted-spend, profitable-harvest, product-target, protected-term, and attribution-incomplete views; and
• implementation notes and next-review conditions, clearly separated from campaign changes.

If thresholds are missing, produce descriptive analysis and request the missing inputs rather than inventing action cutoffs.

Boundaries

Never create keywords, negatives, targets, or bid changes without explicit authorization. Do not assume a negative keyword match type, placement adjustment, or bid percentage from a search-term row alone. Search-term reports contain terms with at least one attributed ad click and may include inferred matches in non-search contexts; absence is not evidence of zero demand. Do not merge currencies or attribution windows, and do not claim incrementality or organic impact from this report alone.

Use [references/test-cases.md](references/test-cases.md) when evaluating future changes.