Amazon negative keywords: how to find them and keep them clean

There is a kind of ad spend that never shows up on any dashboard because it does not register as an error: clicks that were properly charged, on searches genuinely related to your product, that were never going to buy. The shopper who types “free iPhone 15 case” clicks your ad, looks at the price, leaves. You paid. The shopper looking for a spare part when you sell the complete product does the same. So does the one looking for the size you do not carry.

None of those clicks is a system failure. The campaign did exactly what you asked: find related traffic. The problem is that “related” and “buyer” are not the same thing, and the gap between the two is paid at CPC rates every single day. In an account that has been running for months without maintenance, that leak is not a rounding error: it is a slice of the budget going to traffic that is structurally incapable of converting.

Negative keywords are the mechanism for closing that leak. They are not an elegant optimization or an advanced technique; they are hygiene. And like all hygiene, their value is not in doing it once but in doing it always. This article explains where the data comes from, which negative match types exist, why they genuinely lower ACoS, and how to build a maintenance routine that survives contact with a busy week.

One warning first: the Amazon Ads interface changes often. Menu names, report locations and per-campaign limits get adjusted periodically. What follows describes how it works today according to documentation and public 2026 guides, but before you operate, confirm each step in your own console, because the version you see is the one that counts.

where the data comes from: the search term report

The keyword you bid on and the term the shopper typed are not the same thing. That distinction is the heart of the matter. You bid on “school backpack” in broad match; people type “school backpack with wheels”, “cheap school backpack”, “genuine Nike school backpack”, “school backpack repair”. Amazon decides which of those searches resemble your keyword closely enough to show you. The search term report lifts the veil and tells you exactly what the people who saw and clicked your ad actually typed.

The usual route to it, according to guides published in 2026, runs through Seller Central or the Amazon Ads console: go to Advertising, then the campaign manager, then the measurement and reporting section for sponsored ads. There you create a new report, choose the campaign type — Sponsored Products, for instance — select “search term” as the report type, decide whether you want it summarized or with daily detail, set the date range, and download. Repeating the warning: those menu names have moved several times and may look different when you get there.

Two operational details matter a lot. The first is the lag: Amazon Ads data typically carries roughly a 48-hour delay, so yesterday’s searches are not there yet. If you pull the report expecting to see the effect of a change you made last night, you will find nothing and draw the wrong conclusion. The second is the window: a 14- or 30-day range gives far better material than a three-day one, because with little data every term looks like an anomaly.

which negative match types exist

There is an asymmetry here that trips a lot of people up. Positive keywords have three match types — broad, phrase and exact. Negatives only have two: negative phrase and negative exact. There is no negative broad, and that absence is deliberate: a broad negative would be destructive enough to shut down entire campaigns by accident.

Negative exact blocks your ad for that specific search and its close misspellings. It is the scalpel. You use it when you have identified a concrete term that bleeds you and you do not want to touch anything else. Block “school backpack repair” as negative exact and you still show on “repair of school backpack” and any other variation, because you only closed that one door.

Negative phrase blocks any search that contains that sequence of words, whatever else surrounds it. It is the hammer. Block “repair” as a negative phrase and every search including that word disappears at once. Powerful and dangerous in equal measure: a badly chosen negative phrase can switch off profitable traffic you did not know you had.

Public 2026 guides also document limits worth knowing before you build large lists. Amazon allows up to a thousand negative keywords per ad group, and those are additive to whatever you set at the campaign level, so one ad group can accumulate around two thousand exclusions across both levels. On length, negative phrase accepts up to four words and eighty characters, and negative exact up to ten words within the same eighty. As always: verify in the console, because these are exactly the kind of parameter Amazon adjusts without announcement.

Negative product targeting also exists, blocking your ad on specific ASIN detail pages or brands. It is the equivalent for campaigns serving on product pages rather than search results, and it is managed separately from negative keywords.

campaign or ad group: where each negative belongs

The level decision is not administrative trivia, it changes the blast radius. A negative set at the campaign level applies automatically to every ad group inside that campaign. A negative set at the ad group level only affects that group and leaves the rest untouched.

The practical rule that falls out of this is simple. Terms you would never, under any circumstance, want to pay for — “free”, “used” if you sell new, competitor brands you have already proven do not buy from you — belong at the campaign level. Terms that are garbage for one group but legitimate for another belong at the group level. If you have one ad group dedicated to “backpack with wheels” and another to “backpack without wheels”, then “with wheels” is a negative in the second and a star keyword in the first.

The classic mistake is putting everything at campaign level because it is faster. It works for the first few months, and then you discover an entire campaign stopped appearing in a segment that was converting, and nobody remembers who added that exclusion or why. Documenting the reason behind each negative — even as a note on the side — earns its keep the moment someone else has to review the account.

Glossary: ACoS is ad spend divided by the sales attributed to those ads; negatives lower it by removing spend from the numerator without touching the sales in the denominator.

why negatives genuinely lower ACoS

ACoS is spend over attributed sales. Almost every lever for lowering it moves both halves at once: cut the bid and you spend less, but you also sell less, so the percentage barely budges. Raise the price and you improve attributed revenue per click, but you may lose conversion. Almost nothing acts on a single side of the fraction.

Negatives are the exception, and that is why they are the cleanest lever available. When you block a term that piled up clicks and zero sales, you remove spend from the numerator without touching the denominator. The budget that term consumed does not vanish: it gets redistributed toward terms that do buy, because you still have the same daily budget and now there is less garbage competing for it. ACoS falls along two paths at once, and the second — the redistribution — is usually the bigger of the two.

There is a third effect, slower and quieter. Your ad’s accumulated relevance improves when you stop appearing in searches where nobody clicks you. An ad that only shows in contexts where people actually click sustains a higher CTR, and a higher CTR tends to cheapen the click over time. That effect is invisible in week one; it shows up over a quarter.

An example with invented but coherent numbers. Suppose a campaign with 24,000 pesos of monthly spend and 80,000 in attributed sales: a 30% ACoS. Reviewing the report, you find 4,000 pesos of that spend went to eight terms with more than twenty clicks each and no sales. You block them. Campaign spend does not fall to 20,000 because the budget reallocates, but now those 4,000 run through terms converting at the same rate as the rest: roughly 13,000 pesos of additional attributed sales. The campaign lands at 24,000 spend over 93,000 in sales, a 25.8% ACoS. The arithmetic is an illustration, not a promised outcome, but the mechanism is exactly that.

the criteria for deciding what to block

The basic filter for the whole exercise is “clicks without sales”, but applied raw it produces disasters. A term with three clicks and zero sales proves nothing; it may have a perfectly healthy conversion rate and simply not enough history. Blocking it means cutting good traffic out of impatience.

The honest threshold comes from your own conversion rate. If your historical rate is, say, 8%, then a term needs roughly 12 or 13 clicks before it is reasonable to expect a first sale. With fewer clicks than that, zero sales is the most likely outcome even for an excellent term. The practical rule that follows: wait for at least twice as many clicks as the inverse of your conversion rate before declaring a term dead. At 8% conversion, that is about 25 clicks.

With that threshold in hand, candidates split into three piles:

• Block without hesitation. Terms that are another product, another brand you do not carry, spare parts when you sell the whole item, or incompatible intent: “free”, “manual”, “repair”, “second hand” if you sell new.
• Lower the bid, do not block. Relevant terms that do convert but at too high an ACoS. Here the problem is the price of the click, not the fit. Blocking them throws away sales.
• Let it run and watch. Relevant terms with too few clicks so far. They go on a watch list and get reviewed next month.

There is a fourth category almost nobody considers: terms that sell well but for a SKU you cannot supply. Advertising a sold-out size means paying for clicks that land on a listing without a buy button, or on a long delivery promise. There the negative is temporary, and the decision is not an advertising decision but an inventory one.

Glossary: real available stock is sellable inventory net of reservations and in-transit units; advertising terms for a SKU with no real availability is the most expensive way to buy traffic that cannot buy.

the maintenance routine that actually survives

A heroic three-hour cleanup once a year is worth nothing: new terms appear every week because what people search for changes every week. What works is a small, constant routine. Here is one that holds up:

• Every week, fifteen minutes. Pull the search term report for the last 14 days. Sort by spend descending. Look only at the top twenty. Block the ones that clear the clicks-without-sales threshold and are clearly incompatible. Nothing else.
• Every month, one hour. Thirty-day report. Review last month’s watch list and decide on each term. Also review the negatives you added more than 90 days ago: some no longer make sense because your catalog changed.
• Every quarter, half a day. Structural review. Which negatives should move from group to campaign? Which should move down? Is any group’s list close to full and in need of reorganizing?
• Before every peak season. Hot Sale, Buen Fin and December bring searches that do not exist the rest of the year, and they also make normally marginal terms profitable. Review your negatives before the peak starts, not during it.

Two warnings about pacing. First: because of the roughly 48-hour lag, never evaluate the effect of a negative before three or four days have passed. Second: do not block more than a handful of terms per session in a campaign you are still learning to read. Add forty exclusions at once and, if performance shifts, you will not know which one did it.

how all of this looks in iqseller

The search term report tells you which clicks did not convert, but it does not tell you which of the converting ones actually left money behind. That is the missing half, and it is what decides whether a negative is correct. A term at 40% ACoS can be excellent on a wide-margin product and ruinous on a thin one; the same percentage, two opposite verdicts. Without the SKU’s margin beside it, the cleanup happens blind.

In iqseller the Profitability module assembles that context: net margin per SKU with the COGS you loaded, commissions from the Amazon settlement and from MercadoLibre orders, FBA and Full fees, shipping, advertising, and the VAT and withholding breakdown, all along the Parent → Model → SKU tree. Next to it, the Inventory module shows that same SKU’s real available stock, and Alerts flags when it drops, which is precisely the signal to pause terms for something you cannot supply.

With that, the weekly negatives session changes character. You stop asking “did this term spend without selling?” and start asking “did this term spend more than this product can afford?”. The first question finds the obvious garbage; the second finds the spend that looks healthy in the ads console while it quietly eats the margin of the whole SKU. Negatives remain hygiene, but hygiene that knows what everything it is cleaning is worth.