Search Query Performance & Keyword Harvesting

A search term is evidence, not automatically a keyword

Amazon's reports describe different layers of the same demand system. Search Query Performance shows how shoppers move from query impressions toward purchases. Advertising search-term reports reveal which auction matches consumed spend. Targeting reports show how promoted keywords perform after they are given their own bids and match types.

The harvesting decision happens where those layers meet. A term should graduate into a controlled target when it demonstrates relevance, economic viability, and enough repeatable evidence. It should become negative when the account has paid enough to establish mismatch or unacceptable economics—not simply because it failed once.

Three reports, three different questions

Query-level marketplace data gives context; ad reports give controllability. Neither is sufficient alone. High search volume without brand relevance can waste budget, while an efficient low-volume term may deserve protection even if it never appears near the top of a category list.

The brand gained share as shoppers moved deeper into the funnel

Across an anonymized 1,000-query monthly Search Query Performance export, the brand represented 2.54% of 403.2 million marketplace impressions, 3.19% of 6.88 million clicks, 3.24% of 1.82 million cart adds, and 3.72% of 737,810 purchases. The increasing share does not prove advertising caused the purchases, but it does show that the brand's relative performance strengthened after the impression stage.

Report boundary: this funnel is one brand-level monthly export. It is separate from the ad-group search-term and exact-target snapshots used below and is not combined with them to claim incremental lift.

The long tail consumed most of the tested spend without orders

The 221-row search-term snapshot generated 2,868 impressions, 364 clicks, 23 orders, and $471.33 in attributed sales from $241.93 in spend. That implies a 12.7% click rate, 6.3% click-to-order conversion rate, and 51.3% ACoS for this ad-group export. Only 18 distinct terms converted. Non-converting terms accounted for $159.05—65.7% of the spend in this specific snapshot.

Search-term spend by observed order outcome

USD; one Sponsored Products ad-group search-term export dated November 11, 2024; n=221 terms

Bars use total sample spend as the 100% denominator. “With orders” and “without orders” are mutually exclusive term groups and sum to total spend.

This does not mean every non-converting term should be blocked. Some terms have only one click; others may be relevant discovery traffic. The useful insight is concentration: the account needs explicit thresholds so repeated weak signals do not remain invisible inside aggregate ACoS.

From discovery to controlled allocation

The score is a decision model, not a universal Amazon formula. The weighting should change with price, margin, conversion rate, click cost, launch stage, and the account's tolerance for exploration.

Match type should express confidence and control

Set thresholds from break-even conversion, not habit

If a product can contribute $8 per order and the expected CPC is $0.80, the traffic needs roughly a 10% conversion rate to break even before broader strategic considerations. That gives the analyst a defensible way to judge whether ten clicks without an order are acceptable, premature, or already expensive.

A separate targeting snapshot illustrates why controlled targets matter: 136 exact targets produced 992 clicks, 147 orders, $2,965.52 in sales, and 31.4% ACoS. Because this is a different product/report snapshot, it is not a causal comparison with the search-term sample. It does show the richer measurement available once each target has its own history.

Harvesting is a closed loop, not a cleanup task

Source material curated for this framework: Amazon Sponsored Products search-term and targeting exports, Search Query Performance Brand View, Search Catalog Performance, and Top Search Terms reports.

The goal is not more keywords. It is better query ownership.

A mature harvesting system keeps discovery active while steadily improving control. Broad and automatic targeting learn; exact and phrase structures allocate; negatives prevent waste and overlap; query-performance reports reveal whether advertising gains are translating into stronger marketplace presence.

The best outcome is not the largest keyword list. It is a portfolio in which every target has a role, every promotion has evidence, and every negative reflects a documented decision.

Read the metrics as a sequence of decisions.

1. Establish what each dataset can answer

The 1,000-query Search Query Performance export describes the brand’s presence within a marketplace funnel. The 221-row search-term export describes shopper phrases matched inside one Sponsored Products ad-group snapshot. The 136-target report describes separately controlled exact targets from another product/report snapshot. They are useful together as examples of a workflow, but their totals cannot be combined and their efficiency cannot be treated as a before-and-after test.

2. Recalculate the observed search-term funnel

Click-through rate is 364 clicks divided by 2,868 impressions, or 12.7%. Click-to-order conversion is 23 orders divided by 364 clicks, or 6.3%. ACoS is $241.93 spend divided by $471.33 attributed sales, or 51.3%. Each ratio answers a different question: whether the ad earned attention, whether paid visits ordered, and how much advertising cost relative to attributed revenue. None of them alone establishes profit because product margin is not included in that snapshot.

3. Interpret non-converting spend carefully

The $159.05 spent on terms with no observed order is 65.7% of the $241.93 total. That is an allocation signal, not an instruction to negate every row. A relevant term with one click has little evidence; a repeated spender beyond its break-even threshold has more. Review click count, spend, semantic fit, placement, attribution lag, and contribution per order before classifying a term as exploration or waste.

4. Promote only when control adds value

Moving a term to exact match gives it an independent bid, budget context, and performance history. It does not make the query profitable. Verify that the promoted target is serving, then use an exact negative in the discovery source if the routing strategy requires it. Negating too early can suppress useful discovery; failing to negate after verified promotion can make campaigns compete and obscure measurement.

5. Use economics to choose the evidence threshold

Break-even ACoS is contribution margin divided by revenue. Break-even conversion is CPC divided by contribution per order. In the illustrative $0.80 CPC and $8 contribution example, one order per ten clicks is required to cover acquisition cost, so the break-even conversion rate is 10%. This is an example of the decision method, not a disclosed margin for the archived advertising samples.

Why weak aggregate efficiency may not have one cause.

A query decision should leave an audit trail.

The process is complete only when the account can explain why a query sits in its current campaign, what evidence would change that decision, and when it will be reviewed again.