Amazon Keyword Harvesting: Qualify Search Terms, Create Targets and Verify Delivery

Quick answer: qualify the opportunity, then verify the destination

Amazon keyword harvesting means using observed advertising search-term performance to identify terms worth targeting more deliberately. A useful output is not just a keyword list. It is a reviewed mapping from a source term and advertised product to a destination target, matching option, bid and budget context, with a named person responsible for verification.

Start with a validated search term report. Check relevance, evidence maturity, uncertainty, business fit and destination readiness. Create or review the appropriate target through an authorized process, confirm the accepted configuration, then inspect its delivery. Treat any restriction on the original source as a separate decision. A recorded purchase can justify investigation without proving that a term should be promoted immediately.

Amazon has a native Target Promotion feature as well. Its July 2024 announcement describes moving from broader targeting toward more focused targeting—including manual broad match as an example—alongside bid and budget changes. Harvesting is therefore not inherently an exact-match-only procedure. Amazon Target Promotion announcement

Separate search terms, keywords and product targets before moving anything

A report observation is not the same as a configured target

The term you observe and the target you create have different roles. Keep that distinction visible in the review record rather than labeling both columns “keyword.”

Swipe horizontally to view all table columns.

Amazon's search term report can include inferred non-search terms and ASIN-related entries. It includes click-qualified terms, not every impression or a complete transcript of shopper searches. Its documentation also describes promoting keyword and product opportunities. Amazon Sponsored Products search term report

Keyword harvesting is narrower than general keyword research

General research can generate ideas from category language, product attributes or external discovery. Harvesting starts from observed account evidence and asks whether an opportunity deserves a defined targeting action. A potentially relevant phrase found during research has not automatically demonstrated the same thing as a term with attributable advertising activity.

That does not mean harvested terms are guaranteed winners. The source period, product, offer and targeting context still matter. Keep the distinction between “worth testing” and “supported by repeat observations” instead of assigning every candidate the same status.

Product opportunities should not become keyword strings by accident

If the source row represents a product identifier, examine it as a product-targeting candidate. A successful product-page association does not become a keyword merely because it appears in a search-term column. Preserve the original targeting context and verify the identifier before choosing a destination.

For source preparation, use the Amazon PPC search term report analysis workflow. That page covers the evidence and calculation checks; the decisions below begin when the candidate can be traced back to a usable record.

Apply five qualification checks instead of a universal order threshold

1. Relevance and product mapping

Name the specific advertised product that would serve the candidate. Check that it can meet the implied need, including important size, material, compatibility or use-case requirements. Do not promote a term to every product in a campaign because it worked for one item.

The destination may need the same offer or a separately justified alternative. If the advertised item changes, the original performance is evidence from a different product context. Make that difference part of the decision rather than hiding it behind the same keyword text.

2. Comparable and sufficiently mature evidence

Record the account, marketplace, currency, reporting period and attribution definitions. Missing sales information is not a recorded zero. Recent activity may not yet support a stable comparison: Amazon states that attribution metrics remain incomplete until the applicable lookback window ends. Amazon campaign attribution

Do not treat a report downloaded today as entirely mature simply because the file is new. The relevant question is whether the underlying activity has had the appropriate observation period. If the evidence cannot be compared, fix it before ranking candidates.

3. Repeatability and uncertainty

One purchase may reveal relevant intent, but it gives a different level of confidence from repeated observations under comparable conditions. Look for concentration: did the apparent success depend on one short interval, one promotion, or a product situation that no longer applies?

There is no universal number of orders or clicks that this guide can responsibly prescribe for every catalog. A team can define review thresholds, but should label them as operating rules tied to its objective and tolerance for uncertainty—not Amazon requirements or proof of future performance.

4. Business objective and economics

Clarify whether the proposal is for controlled exploration, more deliberate management of established demand, or a different advertising objective. “It converted” is not a complete budget rationale. Review the relevant cost and sales definitions alongside the business's approved economics, without equating attributed revenue with profit.

Historical average CPC is an observation, not a mandatory starting bid. Likewise, a source ACoS is not a guaranteed destination ACoS. Use those figures as context for a reviewed proposal, rather than applying an unexplained bid multiplier or automatically expanding budget.

5. Destination readiness

Before authorizing promotion, identify the destination campaign and ad group, its targeting type, advertised product, status and intended controls. Check whether the same target already exists there. An opportunity may qualify for investigation while its proposed destination is not ready to receive traffic.

This last check prevents an operational mistake from being misread as poor keyword quality. A paused destination and an irrelevant query require different responses.

Four candidate examples: promote, monitor, redirect or repair

A fictional insulated-lunch-bag review

Assume an operator is reviewing opportunities for an insulated lunch bag. The following rows describe invented review situations, not measured performance or prescribed qualification thresholds. The product identifier is a label, not a real ASIN to upload.

Swipe horizontally to view all table columns.

The table shows why the same “harvest” action is not appropriate for every row. A valid result can be reuse an existing target, gather more evidence, select a different entity, or repair the destination.

Why candidate one is not automatically a new keyword

Suppose a recurring process checks only whether a term appears in its latest export. It can repeatedly propose the first row even when the destination already contains the intended target. Check the combination of account, marketplace, product context, destination, target text or identifier, and matching option.

The same words in another ad group are not necessarily duplicates. The objective is to prevent repeated creation of the same intended action while preserving legitimate differences in scope. Keep any deliberate overlap explained rather than assuming that all repeated text is an error.

Why candidate four is not a reason to reject the search intent

The fourth row has a readiness issue, not evidence that the shopper's intent is wrong. If the source were restricted first, the team could create a coverage gap and then blame the keyword. Record the blocked operational step and return it to the person who owns that setting.

Do not fabricate a successful outcome for this example. The next evidence would be the actual accepted configuration and subsequent delivery review, not a promise that activating the destination will restore a particular traffic level.

Specify the destination, match type, bid and budget separately

Use a compatible manual destination

Choose an existing suitable manual campaign or create an appropriate new one. Do not assume an automatic campaign can be turned into a manual campaign in place. Amazon's targeting guidance distinguishes those campaign strategies and explains that changing strategy requires a new campaign. Amazon Sponsored Products targeting guide

Also distinguish campaign copying from target promotion. Amazon's campaign-copy help says a Sponsored Products copy does not switch automatic to manual targeting or keyword to product-attribute targeting, and campaign performance history is not carried over. Those are copying limitations, not a specification of every internal behavior of Target Promotion. Amazon campaign-copy guidance

Select a matching objective, not a ritual

An exact-match target may fit a narrow intent-management objective. Phrase or broad targeting may be relevant when the purpose includes deliberate exploration. Decide what you want to learn or control, then choose and document the matching option. Do not infer that a harvested term must always enter all match types or its own single-keyword campaign.

Matching also does not promise that the new target will reproduce only the identical observed query. Amazon's keyword setup documentation describes variations in manual matching. Preserve the distinction between target-level reporting and the actual terms associated with its activity. Amazon keyword targeting setup

Treat spend-related settings as their own approval

State the proposed bid and the destination's budget context explicitly. If no change to budget is intended, record that too. An approval to add a target should not silently become permission to change campaign spending limits or unrelated targets.

Where a native workflow proposes several adjustments, review each one. More control over a target does not make its economics predictable. Keep the initial decision small enough to inspect and assign responsibility for reviewing actual results.

Add targets manually or evaluate native Target Promotion

Manual setup: turn the record into a specific action

1. Preserve the source evidence and the approved destination specification.
2. Confirm that an equivalent intended target has not already been created.
3. Open the correct account, campaign and ad group, then use the relevant keyword or product-targeting controls.
4. Enter the approved target and settings, and inspect the resulting configuration.
5. Record accepted items, rejected items, identifiers and the change time before moving to delivery review.

For keywords, Amazon documents the path through the campaign and ad group to Targeting, Add keywords, match-type selection and saving. Use the current account labels rather than assuming every screen is identical. Amazon keyword targeting setup

Native Target Promotion: inspect the complete proposal

Amazon's July 2024 announcement introduced Target Promotion in the advertising console and described access through search-term ad-group views and the Amazon Ads API. The current search term report help also references the strategy. Confirm the controls available in your own account; an announcement is not a guarantee of an identical present-day screen. Amazon Target Promotion, Amazon search term report

Compare the proposed source, destination, target type, bid and budget treatment with the review record. If the workflow includes changes that were not approved, stop and resolve that difference. A convenient native action still needs an understood business purpose and an inspectable result.

Bulk and connected workflows: reconcile partial success

For an account that offers a suitable bulk operation, use its current template and supported entity fields. This article's review record is not an upload schema. Do not invent API field names or assume that a copied spreadsheet from another campaign type will be accepted.

Keep per-row results. If some targets were created and others failed, retry only after reconciling that state. A file upload or API request being completed is different from every requested change being accepted, and neither alone proves that the destination is serving.

Decide what happens to the source after the destination is checked

Promotion and source exclusion have different purposes

Adding a target creates an opportunity for more deliberate management. Excluding a term from its original source changes where that intent may be served. These decisions are related, but should not be collapsed into a universal “always negate immediately” instruction.

Document whether the source should remain available for discovery, continue alongside a reviewed destination, or receive a specific routing restriction. The answer depends on the account's objective and observed destination readiness. Do not claim that overlapping campaigns necessarily increase your own CPC or that all overlap is harmless; neither conclusion follows just from repeated keyword text.

Check the scope and side effects of any proposed negative

If routing is approved, specify the correct negative entity, matching option and scope. A campaign-wide restriction could affect products beyond the intended source ad group. Negative matching can also involve close variations, so the reviewer should identify relevant intent to preserve. See the Amazon negative keyword guide for the detailed selection and correction process.

Do not pause an entire discovery campaign simply because one term was promoted. That would affect other opportunities and requires its own rationale. Keep a change record that distinguishes the individual routing decision from broader campaign management.

Preserve a correction path

Record the source's previous state and the separately approved change. If the destination is not meeting the intended operating condition, investigate before making further restrictions. Any correction should address both the live setting and the rule or shared list that created it.

Restoring a setting does not recover missed auctions or guarantee a historical volume. Report what was changed and what was observed afterward, without presenting a rollback as a performance guarantee.

Verify the new target before interpreting its results

Six checks when a harvested keyword has no impressions

1. Identity: Confirm the account, marketplace, advertised product and destination IDs.
2. Target compatibility: Check whether the intended keyword or product target and matching option were created in a suitable destination.
3. Accepted state: Inspect the target, ad group and campaign status, dates, and any rejected operation.
4. Retail readiness: Review the advertised offer's eligibility and availability rather than only the keyword text.
5. Operating constraints: Inspect the approved bid, budget conditions and potentially conflicting exclusions; do not change them automatically.
6. Observation period: Confirm that the report and time range actually cover the new target's activity and allow meaningful interpretation.

Amazon's brand-foundation guidance highlights stock and Featured Offer readiness as relevant to ad display. Its keyword setup help also notes that highly specific terms may receive few impressions. Those are reasons to investigate context, not evidence that every non-delivering target needs a higher bid. Amazon retail-readiness guidance, Amazon keyword setup

Keep source evidence and destination results separate

Attach the original report to the decision, but label new-target results with their own identifiers and period. Do not paste historical source conversions into a destination results column. If a target existed previously, distinguish that target's own history from the evidence that prompted this review.

Assess delivery before claiming performance. Receiving impressions is not the same as producing useful purchases, and attributed sales are not proof of incremental profit. Comparisons should disclose meaningful changes in product, offer, period and settings rather than treating them as a controlled experiment.

Review on a cadence that matches the evidence

A recurring review date can help prevent neglect, but it should not force a change when the data is immature. Separate routine monitoring of status and acceptance from slower assessment of conversion performance. Low-volume terms may need continued observation rather than repeated configuration changes.

If delivery remains unexplained, preserve the configuration, timestamps and relevant reports for the account's support process. Avoid repeated blind additions or bid increases that make the original problem harder to reconstruct.

Choose the workflow level that your review process can support

Manual and native workflows provide the starting point

For a small queue, a source report, a review record and the advertising console can be enough. Native Target Promotion may reduce transfer work where it is available, but its proposed settings still need review. Choose based on the amount of work and the clarity of the accepted result, not merely whether a workflow is automated.

The common limitation is handoff quality: an analyst may identify the opportunity while another person owns product context or spending authority. Standardize the decision record before scaling the transfer mechanism.

Scripts and no-code automation can prepare a repeatable queue

Automate file validation, composite-key checks and candidate formatting before authorizing writes. Retain the original values alongside normalized comparison fields. Use a dry run to show the exact intended destination and action, and stop on missing IDs, changed input structure or ambiguous product mapping.

At scale, monitor exceptions and partial successes. A scheduled job should not promote the same candidate indefinitely or expand its scope to resolve an error. A human should own the unresolved queue and any change to the operating rules.

Agent assistance needs evidence and permission boundaries

An agent can help draft candidate explanations or organize the review packet. A fluent explanation is not evidence that a product is relevant, that a destination is viable, or that an account change is authorized. Check those claims against the original data and actual configuration.

Connected execution requires separate verification of permissions, supported operations and current platform requirements. Permission to analyze a report is not permission to spend or modify campaigns. For a broader operating model, read the Amazon PPC automation guide.

Evaluate the OpenMax handoff with a seven-field promotion record

Where the product fits, and where it does not

OpenMax presents itself as a human-and-agent collaboration platform. The relevant workflow to evaluate here is the handoff among the analyst, product reviewer and authorized operator—not an assumed Amazon keyword executor. OpenMax

This article has not verified OpenMax reading an Amazon ad account, creating targets, setting bids or applying source negatives. Keep those actions in the currently authorized tool unless the required integration and controls have been separately demonstrated. A single operator handling a few candidates may not need an additional collaboration platform.

Start with a read-only review specification

Use this fictional record to check whether another person can reconstruct the intended decision. It is not an Amazon bulk-file format or a depiction of a verified OpenMax interface.

1. Source: account, marketplace, source ad group and report period
2. Candidate: search-term text or verified product-target identifier
3. Evidence: product fit, observations, uncertainty and review rationale
4. Destination: campaign, ad group, advertised product and match choice
5. Controls: proposed bid, budget treatment and approving owner
6. Source treatment: unchanged unless a separate routing action is approved
7. Verification: accepted IDs, change time, delivery check and correction owner

Give the reviewer access only to the information required for the task. Do not put credentials or unnecessary customer information into the packet. Ask the receiving operator to explain what will change, what will remain unchanged, and which evidence would stop the process.

If the read-only handoff works, evaluate a narrow, separately authorized implementation path. Track candidate reviewed, target accepted, delivery checked and source routing reviewed as different states. None of those status labels is a substitute for measuring business results.

FAQ: Amazon keyword harvesting decisions

What is Amazon keyword harvesting?

It is the process of using observed search-term performance to identify and configure more deliberate targeting opportunities. A complete workflow includes qualification, destination specification and verification. Product-target opportunities require their own entity handling rather than being copied into keyword fields.

How many orders or clicks are needed before harvesting a term?

There is no universal threshold suitable for every product and objective. Review relevance, mature comparable evidence, uncertainty, economics and destination readiness. A single purchase can justify investigation or a controlled test without proving repeatable performance.

Should harvested terms always use exact match?

No. Match choice should reflect the intent-management or exploration objective. Amazon's Target Promotion announcement includes manual broad match as an example. Do not assume that harvesting requires exact match, all match types, or a separate campaign for every term.

Can I harvest ASIN opportunities as well as keywords?

Yes, Amazon's report documentation describes product-target opportunities. Verify the ASIN and source context, then use a compatible product-targeting destination. An identifier appearing in the search-term report is not automatically a keyword to upload.

Must I immediately negate the term in the automatic campaign?

Treat source restriction as a separate routing decision. First inspect the intended destination and its delivery, then review the purpose, scope and side effects of any negative. Do not assume that creating a target means its traffic has already moved.

Why does my harvested keyword receive no impressions?

Check destination identity, target compatibility, accepted and active settings, retail readiness, operating constraints and report timing. Low delivery alone does not prove irrelevance or justify an automatic bid increase. Preserve the evidence if the account's support team needs to investigate.

How often should I review harvesting candidates?

Use a recurring review cadence appropriate to activity and team capacity, but separate status monitoring from conversion-performance assessment. If the evidence is still immature or sparse, monitoring can be the right outcome. A calendar reminder does not require a campaign change.

Does harvesting transfer campaign history or guarantee better results?

Do not treat source performance as the new target's results. Preserve it as supporting evidence and track destination activity separately. Amazon explicitly says campaign copying does not carry performance history; target creation also is not proof of improved delivery, profit or organic rankings.

Next step: promote one reviewable decision, not an entire export

Choose one candidate from a validated report and complete the seven-field record. Confirm that the destination is appropriate and that the approving person understands the proposed controls. Leave the source-treatment decision explicit rather than embedding an automatic negative in the transfer step.

If several people share that work, evaluate the read-only handoff with OpenMax before considering a connected executor. The goal is a traceable decision with an accepted setting and a meaningful follow-up—not a larger keyword portfolio that nobody can explain.