Amazon negative keywords tell the ad auction which searches you don’t want to pay for. Most accounts we audit either have almost none, or hundreds added one search term at a time, by feel, with no rule behind them.
Both cost money. The first pays for irrelevant clicks. The second quietly blocks searches that would have converted. Below: how negatives work as of September 2026, then our method — n-gram analysis, a threshold derived from your own margins, and cross-negation between campaigns.
What are Amazon negative keywords?
A negative keyword stops your ad from showing on shopper searches that match it, in the campaign or ad group where you add it. Amazon’s own Sponsored Products guide describes negative targeting as a way to “exclude the keywords, products or brands you don’t want your ads associated with.” It works in both automatic and manual campaigns.
There are three kinds of exclusion in Sponsored Products:
| Exclusion type | What it blocks | Where you can add it (SP, as of Sept 2026) |
|---|---|---|
| Negative exact | Searches that are the keyword, or a close variant of it | Auto and manual keyword campaigns; campaign or ad group level |
| Negative phrase | Any search containing the phrase, in order, or a close variant | Auto and manual keyword campaigns; campaign or ad group level |
| Negative product targeting | Ads showing on specific ASINs, or on a whole brand’s product pages | Auto campaigns and product-targeting ad groups |
There is no negative broad match in Sponsored Products. Your blunt instrument is negative phrase.
Negative exact vs negative phrase: which should you use?
Use negative exact when one specific search is the problem, and negative phrase when a word or short phrase is the problem wherever it appears. Negative exact on “glass water bottle” blocks that search and its close variants only. Negative phrase on “glass” blocks “glass water bottle”, “glass bottle with sleeve” and every other search containing the word.
The limits reported by most PPC tool vendors and agencies, as of September 2026:
| Negative exact | Negative phrase | |
|---|---|---|
| Maximum words | 10 | 4 |
| Maximum characters | 80 | 80 |
| Blocks | The exact search (plus close variants such as plurals) | Any search containing the phrase in sequence |
| Risk | Low — very narrow | High — one word can block hundreds of searches |
| Best for | Cross-negating, one-off bad terms | Irrelevant attributes, wrong buyer, wrong product type |
Amazon’s library guide says both negative types also block “close variations” of the phrase. In practice that means plurals: negating “straw” also blocks “straws”. Guides disagree on whether misspellings are caught, so if a misspelling is costing you money, add it as its own negative.
The mistake we see most often: negative phrase used where negative exact was meant. Someone sees “water bottle for kids” wasting money, adds “water bottle” as negative phrase, and the campaign stops serving on half the category.
Should negatives go at campaign level or ad group level?
Put a negative at campaign level when it should never trigger anything in that campaign, and at ad group level when it’s only wrong for one ad group. Campaign level negative keywords apply to every ad group in the campaign. Ad group negatives apply only to that ad group.
Our rules of thumb:
- Campaign level: wrong product type (“glass”, “plastic”), wrong buyer, competitor brand names in non-conquest campaigns, and anything that fails the threshold across the whole campaign.
- Ad group level: cross-negation (below), and terms that are wrong for one product variant but right for another sitting in the same campaign.
The risk with campaign-level negative phrase is collateral damage. If “kids” is a campaign negative and one ad group targets “kids water bottle 500ml”, that ad group is now dead. Search your keyword list before adding one.
How does negative product targeting work?
Negative product targeting stops your ads showing on specific product detail pages, either by ASIN or by excluding a whole brand. You’ll find it on the “Negative targeting” tab of auto campaigns and product-targeting ad groups, as of September 2026. Amazon’s guide suggests excluding low-performing ASINs, ASINs you don’t want your ads next to, and non-complementary products.
Two uses matter most:
- Auto campaigns. The “substitutes” and “complements” groups will place your ad on pages that convert terribly. Filter the auto search term report to ASIN rows and apply the same threshold as for keywords.
- Your own ASINs. In competitor-conquesting campaigns, exclude your own brand so you aren’t paying to advertise one product on another.
Sponsored Brands also supports negative exact, negative phrase and negative product targeting; Sponsored Display supports negative product targeting. The method below works across all three, but we start with Sponsored Products because its search term data is the richest.
What changed in 2025–2026?
The mechanics of Amazon negative keywords have barely moved; what’s changed is the search terms they have to deal with. With Rufus and conversational shopping handling a growing share of queries, search term reports now show longer, sentence-like searches. A single negative exact on “water bottle that keeps drinks cold for 24 hours” will almost never fire again. Word-level analysis matters more than it did three years ago.
There were reports in late 2025 of Amazon adding negative keyword options to product-targeting ad groups. We couldn’t confirm this against an Amazon announcement, so check your own console before relying on it. We also found no account-level or portfolio-level negative keyword list for Sponsored Products as of September 2026. Negatives are still added campaign by campaign — bulk sheets are the realistic way to apply one list across many campaigns.
What is n-gram analysis on Amazon?
N-gram analysis breaks every search term into its individual words (1-grams), word pairs (2-grams) and word triples (3-grams), then adds up clicks, spend, orders and sales for each one. Instead of judging 2,000 search terms with three clicks each, you judge 300 words with enough data to decide.
Take the search term “insulated water bottle kids”. It produces:
- 1-grams: insulated, water, bottle, kids
- 2-grams: insulated water, water bottle, bottle kids
- 3-grams: insulated water bottle, water bottle kids
Do that for every row in your search term report, give each n-gram the full metrics of every search term it appears in, then group and sum. A pivot table does it. We run it on 60 days of data, per product, never per account — “kids” is waste for one product and the best word you have for another.
N-gram analysis on Amazon finds what term-by-term review can’t: a word losing money in small amounts across dozens of different searches. Our broad match drift post shows how quickly those small amounts build up.
How do you decide when to negate a term?
Negate an n-gram or search term when it has had enough clicks that zero orders is unlikely to be bad luck, and it has spent more than an order would be worth to you. Both parts come from your own numbers. This is our heuristic, not an Amazon rule.
Start with the maths from break-even ACoS:
- Break-even ACoS = profit per unit before ad spend ÷ selling price
- Break-even CPA = selling price × break-even ACoS (the most you can pay for an order and not lose money)
- Target CPA = selling price × target ACoS
- Expected clicks per order = 1 ÷ your conversion rate (CVR)
Then the rule:
Negate when: orders = 0 and clicks ≥ 2 ÷ CVR and spend ≥ 1.5 × target CPA
Negate or cut the bid when: orders ≥ 1 and ACoS > break-even ACoS, with clicks ≥ 2 ÷ CVR
Why 2 ÷ CVR and not 1 ÷ CVR? At 1 ÷ CVR clicks, a term converting at exactly your average rate still gets zero orders about 37% of the time. That’s too often to act on. At 2 ÷ CVR, the chance drops to about 13%. The probability of zero orders by chance is (1 − CVR)^clicks, so you can set the bar where you like. At 3 ÷ CVR it’s about 5%.
The 1.5 × target CPA check stops the rule firing on cheap clicks where waiting costs little.
Worked through, with illustrative figures: a £24 bottle, 40% margin before ads, 25% target ACoS, 12% conversion rate.
- Break-even ACoS = 40%; break-even CPA = £24 × 0.40 = £9.60
- Target CPA = £24 × 0.25 = £6.00
- Click threshold = 2 ÷ 0.12 = 16.7, so 17 clicks
- Spend threshold = 1.5 × £6.00 = £9.00
A term with 17 clicks, £9+ spend and no orders gets negated. On a £120 product converting at 6%, the click threshold becomes 34. Expensive products need more patience, not less.
A worked n-gram example (illustrative)
Here’s what 60 days of n-gram output looks like for a made-up product — a 1-litre insulated steel water bottle at £24. The figures are invented to show the method. They’re not client data. Thresholds from above: 17 clicks, £9.00 spend, 40% break-even ACoS.
| N-gram | Type | Search terms containing it | Clicks | Spend | Orders | Sales | ACoS | Decision |
|---|---|---|---|---|---|---|---|---|
| glass | 1-gram | 14 | 41 | £32.80 | 0 | £0 | — | Negative phrase, campaign level |
| plastic | 1-gram | 9 | 19 | £14.44 | 0 | £0 | — | Negative phrase, campaign level |
| kids | 1-gram | 22 | 63 | £47.25 | 2 | £48 | 98% | Negative phrase (wrong buyer for a 1L bottle) |
| tumbler | 1-gram | 12 | 31 | £24.80 | 1 | £24 | 103% | Negative phrase in broad; test separately if strategic |
| pink | 1-gram | 5 | 8 | £6.40 | 0 | £0 | — | Below threshold: wait |
| with straw | 2-gram | 11 | 27 | £21.60 | 1 | £24 | 90% | Negative phrase if no straw lid |
| insulated bottle | 2-gram | 18 | 88 | £70.40 | 13 | £312 | 22.6% | Keep; harvest top terms to exact |
| water bottle 1l | 3-gram | 6 | 34 | £27.20 | 5 | £120 | 22.7% | Harvest to exact, cross-negate |
Look at the “glass” row. Fourteen different search terms, 41 clicks between them. Individually, most of those terms had two or three clicks and would never trip a per-term rule. Together they spent £32.80 for nothing. That’s what n-gram analysis is for.
Now look at “pink”. Eight clicks, no orders, and it’s tempting. But at a 12% conversion rate, eight clicks with zero orders happens by chance roughly 36% of the time. Leave it.
One caution: an n-gram’s numbers overlap. “insulated bottle” and “water bottle 1l” share search terms, so the rows don’t add up to your account total. Judge each n-gram on its own, and before adding a negative phrase, check which of your converting search terms it would also block.
How do you cross-negate to stop cannibalisation?
When a search term graduates from a discovery campaign into its own exact-match campaign, add it as negative exact in every discovery campaign it came from. Without that, your broad, phrase and auto campaigns keep bidding on the same search, and you lose control over which bid and which campaign wins it.
This is the core of the SKAG structure. The flow:
- Auto, broad and phrase campaigns find search terms.
- A term that converts below target ACoS (we look for two or more orders) moves into a single-keyword exact ad group with its own bid.
- That exact term goes in as negative exact — never negative phrase — in the auto, broad and phrase ad groups that found it.
Cross-negate “water bottle 1l” as negative phrase and you also block “water bottle 1l insulated” and every long-tail search discovery might still find. Negative exact closes one door and leaves the rest open.
Amazon doesn’t publish how it picks between two of your own ads eligible for the same search. Cross-negation takes that decision out of its hands. Do the same across sizes: negate “500ml” in the 1L campaigns and vice versa.
What should you never negate?
Never negate words that describe your core product, terms that sell organically, or your brand in brand-defence campaigns — and never negate on thin data. The expensive negatives we remove during audits nearly all fall into one of these:
- Core product words. Negative phrase on “bottle” in a water bottle campaign. We still find it.
- Terms below the threshold. Five clicks and no sale is not evidence. Use the 2 ÷ CVR rule.
- Terms that sell organically. Check Search Query Performance before negating. If a term drives organic purchases, a weak ad ACoS may be supporting rank. Read ACoS vs TACoS before you cut it.
- Seasonal terms, off-season. “Christmas gift for runner” in March looks dead. It isn’t in November. Pause the keyword; don’t negate the phrase.
- Your own brand, in brand-defence campaigns. Negating your brand name in discovery campaigns is sensible. Negating it everywhere hands the search to competitors’ Sponsored Brands ads.
Review the whole negative list quarterly. Add a straw lid, and “with straw” goes from waste to your best new keyword while still sitting in the negatives.
How often should you review Amazon PPC negative keywords?
Monthly, per product, on 60 days of search term data. Our routine for Amazon PPC negative keywords takes about an hour per product:
- Download the Sponsored Products search term report and split out ASIN rows.
- Build 1-, 2- and 3-grams and apply the threshold. Flag, don’t auto-apply.
- Check each flag against your positive keywords and SQP organic data.
- Add survivors at the right level, and cross-negate anything harvested to exact.
- Log every negative with its date and numbers, so nobody wonders in a year why “tumbler” is blocked.
FAQ
What is the difference between negative exact and negative phrase on Amazon?
Negative exact blocks one specific search term and its close variants, such as plurals. Negative phrase blocks any search that contains the phrase in the same word order, so a single word can block hundreds of searches. Use exact for one-off bad terms and cross-negation, and phrase for irrelevant words like the wrong material or buyer.
How many words can an Amazon negative keyword have?
PPC tool vendors and agencies consistently report a maximum of four words for negative phrase and ten words for negative exact, with an 80-character limit on both, as of September 2026. If you need more than four words to describe an exclusion, it’s usually a negative exact job.
Should I add negative keywords at campaign or ad group level?
Campaign level for anything that should never trigger an ad in that campaign, such as the wrong product type or wrong buyer. Ad group level for cross-negation and for terms that are wrong for one variant but right for another in the same campaign. Check campaign-level negatives against every ad group’s keywords first.
How many clicks before I negate a keyword on Amazon?
We use a heuristic of two divided by your conversion rate, with zero orders and spend above 1.5 times your target cost per order. At a 12% conversion rate that’s 17 clicks. Below that, a zero-order result is too likely to be chance. Higher-priced products with lower conversion rates need more clicks before you act.
Can I add negative keywords to automatic campaigns?
Yes. Automatic campaigns accept negative exact and negative phrase keywords, plus negative product targeting for ASINs and brands. Auto campaigns need negatives more than most, because they find search terms and product pages you’d never choose. Cross-negate any term you’ve moved into an exact campaign as negative exact in the auto campaign.
What is n-gram analysis in Amazon PPC?
It’s splitting every search term into single words, pairs and triples, then summing clicks, spend and orders for each. That shows words losing money in small amounts across many different searches, which you can’t see one search term at a time. It’s the most efficient way to build a negative phrase list.
Sami Sultan runs Ecom Enable, an Amazon growth partner for DTC brands. Negative match types, limits and targeting options checked 22 September 2026. If you’d like us to run this negative keyword method on your account, it’s part of our Amazon advertising service, or start with a free audit.
Sources: Amazon Ads: A guide to targeting with Sponsored Products; Amazon Ads: Best practices for Sponsored Products; Amazon Ads help: Negative keyword, product, and brand targeting (page requires sign-in; not readable by us); Amazon Ads API: Sponsored Products negative targeting; Amazon Ads API: Sponsored Brands negative keyword targeting; Karooya on negative phrase vs negative exact limits; Ad Badger on negative keywords; Ad Badger on negative ASINs in auto campaigns; Karooya on negative product targeting; Scale Insights on negative keyword levels; EcomRanker on 2026 search term changes.