Amazon negative keywords: how to decide what to block without cutting good demand

A search term with spend and no orders deserves attention. It does not always deserve a negative keyword.

The decision depends on relevance, the amount of evidence, the age of the data, the product's economics and the consequences of the match type. A simple rule applied without those checks can remove useful demand as efficiently as it removes waste.

Amazon's search-term report shows terms that produced at least one ad click. It can help identify positive and negative targeting candidates, but it is not a record of every possible search impression. Amazon's report documentation.

Check what each row represents. The report can include shopper-entered queries, terms inferred for non-search placements and ASINs representing product-detail-page placements. An ASIN exclusion uses negative product targeting, not a negative keyword.

Separate “wrong shopper” from “not yet profitable”

Start with two buckets.

Clearly irrelevant terms describe a product, compatibility, size or use case you genuinely do not offer. For example, if you sell an unpowered wooden toy, a search explicitly requiring a rechargeable electronic version may be a poor fit. Read the actual listing before deciding: assumptions about a product can be wrong too.

Relevant but weak terms need more diagnosis. A shopper can want the right product and still fail to buy because of price, delivery, unclear images, unavailable variations or an expensive click.

Negation removes eligibility for matching demand. A bid reduction changes the price you are willing to pay for it. A listing fix addresses a different part of the problem. Choose the tool that fits the evidence.

Make the data comparable

Before evaluating a term, record:

• Marketplace, advertised product and campaign.
• Search term and current targeting source.
• Reporting dates and applicable attribution window.
• Clicks, spend, orders, attributed sales and contribution assumptions.
• Any price, promotion, stock or listing change during the period.

Use sufficiently mature data for the reporting setup. Do not compare yesterday's unresolved clicks with a fully developed historical period and label the difference a trend.

For a search term appearing across campaigns, inspect both the combined picture and the contexts. One product can fit the query while another does not. A campaign-wide negative can remove both.

Why a universal click threshold is weak evidence

Amazon's Sponsored Products targeting guide recommends assessing performance after at least 20 clicks before adding a negative on performance grounds. That is useful guidance, not a guarantee that every 20-click sample supports the same conclusion. Amazon's targeting guide.

Consider a hypothetical term with a true 10% conversion probability per click. Under a deliberately simplified model of independent, identical clicks, the probability of seeing zero orders after 20 clicks is 0.9 to the power of 20, or about 12%.

That does not mean you should keep funding it. It means zero orders is compatible with more than one explanation.

A high-priced product with expensive clicks may exhaust a sensible test budget before a large sample arrives. A low-cost click on a very relevant term may justify more learning. Define the maximum affordable uncertainty instead of pretending that one threshold eliminates it.

Match type determines the blast radius

Amazon supports negative exact and negative phrase. Its documentation includes close variations in both descriptions. Do not assume negative exact means only one literal character-for-character query. Amazon's negative-targeting documentation.

Before adding a phrase negative, list the valuable searches that contain that phrase. A root such as “gift” could occur in many relevant queries. Blocking it to solve one weak query can create a much larger problem.

A good review note says:

“Proposed negative exact for this query in this ad group because it requires a feature the advertised product does not have. Reviewed related converting terms. No broader phrase exclusion proposed.”

For a performance-based exclusion, add the reporting period, mature spend, economics and alternative actions considered.

Use a four-way decision

Every reviewed term should end in one of four places:

1. Exclude: clear irrelevance or sufficient evidence against the agreed economics.
2. Reduce exposure: relevant demand, but the current cost is too high.
3. Continue a bounded test: plausible opportunity, insufficient evidence, a defined spending limit.
4. Investigate the offer: the term fits, but conversion deteriorated across several traffic sources.

Do not call all four “optimization.” State the actual decision.

Check the effect after the change

Keep the original setting and the reason in a change log. After enough comparable data accumulates, check whether irrelevant spend fell, useful demand was preserved and total contribution moved as intended.

A lower ACoS after removing traffic is not complete evidence of success. You may also have removed profitable volume. That is why the contribution-profit view matters.

If the problem appears across many relevant searches, use the conversion-versus-traffic diagnosis before adding more negatives.

For an evidence-led review of the account, start here.

‍