From Amazon Search Terms to Controlled Keyword Harvesting

Amazon search term harvesting is one of the most practical, repeatable skills an Amazon seller can develop. Done well, it turns the raw signal inside your auto and broad campaigns into a tightly managed keyword list — one where every term either earns its place or gets excluded. Done poorly, it becomes a cycle of overspending on terms that never convert and under-investing in the ones that do. This guide explains the full workflow: where to find the data, how to read it honestly, and how to make match-type and negative decisions without shortcuts that can quietly damage your campaigns.

What search term harvesting actually means

When Amazon runs an automatic or broad-match campaign, it matches your ad to customer queries it thinks are relevant. Those queries appear in the Search Term Report, which Amazon makes available through Seller Central and via the Amazon Ads API. The report shows you the actual words a customer typed before clicking your ad — not just the keyword you bid on.

Harvesting is the process of reviewing those queries and deciding one of three things:

Promote it — add it as an exact or phrase keyword in a controlled campaign with a deliberate bid.
Leave it — the term is still too new to judge; keep collecting data.
Exclude it — add it as a negative keyword to stop wasting budget on searches that do not convert for your product.

That three-way decision, applied consistently, is the entire discipline.

Where the data comes from

Amazon provides two primary sources sellers can access directly:

Search Term Report (Sponsored Products) — available in Seller Central under Reports → Advertising Reports. It shows impressions, clicks, spend, orders and revenue at the query level.
Amazon Ads API — the same data is programmatically accessible via the GET /v2/reports endpoints documented at advertising.amazon.com/API/docs. Tools that connect to this API can surface the same report data in structured form.

Neither source tells you the total search volume for a term across Amazon. They only show performance for your ads, which means low impressions on a query does not necessarily mean the term is unpopular — it may mean your bid or eligibility is limiting exposure.

Reading the Search Term Report honestly

Before promoting or excluding anything, apply a minimum evidence threshold. A single click with no purchase proves very little. Common thresholds sellers use:

For exclusions: enough clicks to statistically rule out a conversion at your break-even ACoS. If one click costs $0.80 and your average order value makes break-even ACoS 30%, you need roughly 3–5 clicks with zero sales before the signal is meaningful — not one.
For promotions: at least one confirmed order, preferably two or more, before moving a term into a tightly bid exact campaign.

These are not fixed rules. They depend on your price point, margin and how competitive the category is. The point is to decide your thresholds before you open the report, so the data drives the decision rather than cherry-picking.

Match-type decisions explained

Once a term has enough evidence to promote, choosing the right match type controls how broadly Amazon continues to expand from it:

Exact match — Amazon shows your ad only when the customer types that precise query (with minor variations like plurals). Use this for proven, high-intent terms. It gives you maximum bid control.
Phrase match — Amazon shows your ad when the query contains your keyword as a continuous phrase. Use this when a term has proven itself but you want to capture natural extensions (e.g., “running shoes for women” capturing “best running shoes for women”).
Broad match — Amazon expands freely. Best used in dedicated discovery campaigns where you intend to harvest from it, not as a destination for proven terms.

A disciplined harvesting workflow creates a funnel structure: auto/broad campaigns generate discovery; exact and phrase campaigns capture proven demand with controlled bids.

Adding negatives: the other half of the work

Negatives are at least as important as promotions. Every irrelevant search term consuming budget without converting is a cost you can eliminate. There are two levels:

Campaign-level negatives — block a term across all ad groups in a campaign.
Ad group-level negatives — block a term only within a specific ad group.

A common structural use of negatives is query sculpting: once you promote a term to an exact campaign, add it as a negative exact to the auto/broad campaign that discovered it. This prevents the same query from triggering two campaigns competing against each other (which inflates your own costs).

Negative match types work differently from positive ones. A negative exact blocks only that precise query. A negative phrase blocks any query containing that phrase. Be careful with negative phrase — it is easy to accidentally exclude relevant traffic.

Practical harvesting checklist

Use this sequence every time you run a harvest cycle:

Pull the Search Term Report for the period you are reviewing (minimum two weeks; four weeks preferred for lower-volume campaigns)
Set your minimum evidence thresholds before filtering the data
Sort by spend descending — high-spend, zero-order terms are your first exclusion candidates
Sort by orders descending — multi-order terms with acceptable ACoS are your first promotion candidates
For each promotion candidate, decide: exact, phrase, or both?
For each exclusion candidate, decide: campaign-level or ad group-level? Exact or phrase negative?
Check whether any promoted term already exists in your exact campaigns — avoid adding duplicates
Apply query sculpting: add newly promoted exact terms as negative exacts in the source campaign
Document every change — the date, the term, the action and why

Tools like AdsPilot are designed to structure this checklist as a repeatable workflow, surfacing negative candidates and match-type evaluations from connected account data, then routing proposed changes through explicit approval gates before anything is written to the Amazon Ads API. Changes do not execute autonomously — each action passes through an approval step, and every decision is logged in an audit trail. See the AdsPilot feature overview for more on how that approval model works.

Limitations to understand before you start

Search term data has a delay. Amazon typically reflects recent data with a 48–72 hour lag. Do not make decisions on data from the last two days.
The report only shows terms that received at least one click. You cannot see what Amazon matched you to but did not result in a click.
Attribution windows affect order counts. Amazon’s default attribution window for Sponsored Products is 7 days. A term with clicks but zero orders within your report date range may have orders attributed outside that window.
Marketplace data is separate. Each Amazon marketplace (US, UK, DE, etc.) has its own search term data. Harvesting decisions are not transferable across marketplaces without reviewing local data.
API access requires granted roles. Programmatic access to search term data via the Amazon Ads API requires that the appropriate advertiser permissions be granted to the connecting application. Review the Amazon Ads API documentation for the current scope requirements.

For sellers working across multiple accounts, AdsPilot’s tenant-bound security model is designed to ensure that data and actions from one account cannot affect another — a relevant consideration when multiple team members or agencies share access.

Sources

Amazon Ads API documentation: https://advertising.amazon.com/API/docs
Amazon Seller Central Advertising Reports (accessible via your Seller Central account)
AdsPilot Amazon Ads API integration

This article describes the intended completed AdsPilot workflow. Its practical scope depends on the connected Amazon account, marketplace, granted API roles and the evidence those sources actually provide.