A negative keyword blocks your ads from showing on a search term you name. That is the whole mechanism, and it is the most underused lever in Amazon PPC. In the accounts I audit, sellers will tune bids weekly for a year while the same unconverting search terms take clicks every day, unchallenged. This guide covers what negatives actually do, where legitimate candidates come from, the negative exact versus negative phrase decision (the part sellers get backwards most often), how many clicks a term deserves before you pass sentence, whether the negative belongs at the campaign or ad group level, and the mistakes that quietly cost more than the bleed they were meant to stop.
What a negative keyword is, and the two jobs it does
A negative keyword is an instruction that stops your ads from appearing on specific search terms. Regular keywords tell Amazon where you want to show; negatives tell it where you refuse to. Sponsored Products supports two match types, negative exact and negative phrase. There is no negative broad. (Negative product targeting exists too, for ASINs, but this guide is about keywords.)
Negatives matter because most of your traffic is traffic you never chose. Broad match, phrase match, and auto targeting fan one keyword out into hundreds of search terms, and as I covered in my search term report guide, the only control you have over those searches is negatives and harvests. In practice negatives do two jobs: they stop you paying for searches that never convert, and they enforce ownership, so that after you harvest a term into exact match, the discovery campaign that found it stops re-buying it and splitting the data.
Where negative keyword candidates come from
Every negative worth adding comes out of the search term report: real queries that took your clicks and gave nothing back. The primary source is zero-order spend. The secondary source is high-ACOS terms that still convert, and that one is a judgment call which usually ends in a bid cut instead.
Open the report, filter orders to zero, sort by spend descending, and the candidates introduce themselves. How much zero-order spend is tolerable depends on phase: my benchmarks guide allows 50–60% of spend at launch, tightening to 25–30% for a mature account, because discovery is an investment early and a leak late.
The free Negative Keyword Finder builds the list for you: upload the same bulk file, and it flags zero-order and high-ACOS search terms against your own target ACOS, recommends exact or phrase for each, and exports the lot in a ready-to-upload format. It parses in your browser, no email, no account.
One warning on the high-ACOS bucket before you negate anything in it: a term that converts above your target ACOS is mispriced, not worthless. Negating it surrenders orders and whatever organic rank those sales were feeding. Most of the time the right verdict is a smaller bid, and I reserve negatives for terms that do not pay at any price.
Negative exact vs negative phrase: when to use each, and why
Negative exact blocks the one search term you name, plus close variants like plurals. Negative phrase blocks every search term that contains your phrase, whatever surrounds it. One is a scalpel, one is a demolition order, and the craft of negation is knowing which the evidence in front of you actually justifies.
Watch the difference on one made-up family. Say your keyword is broad match "laptop sleeve" and the report shows "pink laptop sleeve" bleeding while "laptop sleeve 13 inch" and "neoprene laptop sleeve" convert. A negative exact on "pink laptop sleeve" removes that query and nothing else. A negative phrase on "pink" removes every search containing the word: the bleeder, yes, but also every pink variant that ever converted or ever would. Same intention, radically different blast radius.
The reason exact is my default is that the two mistakes are different sizes. A wrong exact negative costs you one search term, and you can see exactly which one you blocked. A wrong phrase negative deletes a whole family of queries, including converters you never listed, and the damage never appears in any report, because blocked impressions do not generate rows. Negatives fail silently. That asymmetry is the entire argument, and it is why phrase has to earn its way in with two proofs: the pattern fails across many terms, and no converting term contains it.
Both proofs come from the same place. The free N-Gram Analyzer breaks your search terms into word-level patterns. If one word shows up in fourteen failing queries and zero converting ones, that is a phrase negative: block the pattern once instead of chasing fifty variants one exact at a time. If the word appears in even one strong converter, the phrase stays in the drawer.
| The situation | The call | Why |
|---|---|---|
| One search term bleeds; its siblings convert | Negative exact | Kills only the named query. The converting neighbors keep serving. |
| The same word fails across many different search terms | Negative phrase, after the check below | One negative blocks the whole family instead of fifty exacts chasing variants. |
| The failing word also appears inside converting terms | Negative exact, term by term | Phrase would take the converters down with the junk. Slower, but nothing valuable dies. |
| You harvested a term into its own exact match keyword | Negative exact in the source campaign | The discovery campaign stops re-buying a term you already own, and keeps exploring everything else. |
| A term converts, but above your target ACOS | Neither. Cut the bid | Negatives are for terms that do not pay at any price. This one pays at a lower one. |
How many clicks before you negate? Your account already knows
There is no universal click threshold, and any guide that hands you one is guessing with your money. The honest bar is your own conversion rate: one divided by CVR is the number of clicks that buys an average order. A zero-order term earns its verdict against that number, not against a rule of thumb.
Run it on the dataset I grade across these guides: a 9.1% conversion rate means roughly one order per eleven clicks. A zero-order search term approaching that count is not unlucky; measured against the account's own physics, it has had its trial, and every click past it is a bet at longer odds.
The bar moves with the account, which is the point. A 3% conversion account needs around thirty-three clicks for the same confidence, and an eleven-click rule there executes terms before trial. It also moves with phase: at launch, tuition is the plan and trials run long; in the profitability phase, patience is expensive. Whatever your bar, judge it over a 60-day window so slow converters get a fair hearing before sentencing.
Campaign level or ad group level?
Add the negative at the campaign level when the term is wrong for everything the campaign sells, and at the ad group level when it is wrong for one product but a fair match for its neighbors. In a one-product-per-campaign structure the two collapse, and campaign level is the cleaner default.
The one structural case worth memorizing: ownership negatives after a harvest go at the campaign level, in the discovery campaign that found the term. The exact match campaign now owns the query; the auto or broad campaign that surfaced it should never buy it again, and a campaign-level negative exact is what enforces that. This works inside auto campaigns too, where negatives are the only steering you have.
The mistakes that cost more than the bleed
The expensive negation mistakes are all one mistake in different clothes: acting on evidence you do not have. Negating before a fair trial, phrase-blocking a word that lives inside your converters, and killing expensive terms that still convert all destroy revenue quietly, because a blocked search never reports the sales it would have made.
The subtler version is the negative list nobody re-reads. Negatives outlive their reasons: a term fairly executed during launch may deserve a retrial after new images, a new price, or a new season, but no report will ever nominate it, because blocked terms produce no data. And an auto campaign negated into a corner stops minting harvest candidates entirely. Discovery has a cost, but it also has a yield, and over-negating trades both away.
A worked example: the full policy on real numbers
Here is the framework running on the dataset from my benchmarks guide: $36,303 in ad spend, $111,058 in ad sales, 4,140 orders from 45,672 clicks. The toplines compute to a 32.7% ACOS, a 9.1% conversion rate, and a $0.79 average CPC. Three averages, and they contain this account's entire negation policy.
The policy falls out in three steps. First, the verdict bar: at 9.1% conversion and $0.79 a click, a zero-order term nearing eleven clicks has cost one average order's worth of budget, about $8.77 by dividing spend by orders; by twenty-two clicks and roughly $17 it has cost two, and calling that bad luck takes more optimism than I can price. Second, the match type: if the dead term is an outlier in a family that converts, it gets one negative exact and its siblings keep serving. If the n-gram read shows its core word failing across the family with no converters attached, one negative phrase closes the whole account of it. Third, placement: in the campaign that spent the money, at campaign level unless a sibling product has a legitimate claim to the term.
The Negative Keyword Finder runs this exact arithmetic on your own file: your conversion rate, your CPC, your target ACOS, and an exact-or-phrase call per term, exported ready to upload. What it will not do is override your judgment on the high-ACOS bucket, because it should not. That call is yours.
Frequently asked questions
What is a negative keyword on Amazon?
A negative keyword blocks your ads from appearing on specific search terms. Regular keywords tell Amazon where you want to show; negatives tell it where you refuse to. Sponsored Products supports two match types, negative exact and negative phrase. There is no negative broad match.
What is the difference between negative exact and negative phrase?
Negative exact blocks only the search term you name, plus close variants like plurals. Negative phrase blocks every search term containing your phrase, whatever surrounds it. Exact is the surgical default; phrase is powerful but indiscriminate, and it earns its place only when a whole pattern fails and no converting term contains it.
How many clicks before I add a negative keyword?
There is no universal number. The honest bar is one divided by your conversion rate: the clicks that buy an average order in your account. At the 9.1% conversion rate of the dataset in this guide, that is roughly eleven clicks; a 3% account needs closer to thirty-three for the same verdict.
Can negative keywords hurt my organic ranking?
Not directly. A negative only stops your ad from matching a search; it touches nothing on the organic side. The indirect risk is real, though: negate a term that was producing ad sales and you lose the sales velocity feeding your rank on it. That is why converting terms get bid cuts, not negatives.
Do I add negatives at the campaign level or the ad group level?
Campaign level when the search term is wrong for everything the campaign sells, ad group level when it is wrong for one product but a fair match for its neighbors. In a one-product-per-campaign structure the question collapses, and campaign level is the cleaner default: one place to audit later.
Should I negate a high-ACOS search term that still converts?
Almost never. A converting term running above your target ACOS is mispriced, not worthless: cut the bid to what the term earns per click and keep the orders. Negatives are for search terms that do not pay at any price. Reserving them for that protects both the revenue and the data.
Stop paying for searches that never pay you back
The fastest way to size the problem is the free Account Health Snapshot: upload your bulk file and it grades nine metrics, wasted spend included, in about 60 seconds, parsed in your browser with no email and no account. And if your negative list raises a question a scorecard cannot answer, which campaign owns which term, what deserves a retrial, where phrase is safe, the free 30-minute diagnosis call is where I make those calls with you, term by term.