Search term bleed is when multiple campaigns in the same Amazon account buy the same customer search term, usually because broad or phrase match keywords in one campaign match searches another campaign was built to own. The term's conversion data splits across campaigns, so no single campaign has the statistics to optimize it, and its effective ACOS inflates. I coined the name in my audit work because that is exactly how it behaves: not a crash you hear, a leak you find months later. Sellers search for the problem as keyword cannibalization, search term overlap, or duplicate search terms across campaigns. Whatever you call it, it is the structural problem I find most often in accounts that otherwise look well managed, because every individual row still looks plausible. This guide covers why bleed happens, what it costs, how to spot it in your own report, and the ownership fix.
What search term bleed is (and what it is not)
Search term bleed is a search term problem, not a keyword problem. Two campaigns can hold completely different keywords and still buy the same search, because broad match, phrase match, and auto targeting all expand past the words you typed. That is why bleed survives keyword-level audits: the duplication happens one layer below the keywords.
The distinction matters because most sellers police duplication at the keyword level. They confirm no keyword appears twice, call the account clean, and never see that broad match "laptop sleeve" in a research campaign and exact match "neoprene laptop sleeve" in a performance campaign are both buying the search "neoprene laptop sleeve 13 inch." Amazon widens the net further: its exact match also covers plurals, misspellings, and reordered words, so even two exact keywords can collide. That keyword-versus-search-term distinction is where my search term report guide starts; bleed is the fourth of the four things I hunt for there.
Nor is it quite keyword cannibalization, which usually describes ads eating sales a listing already earns organically. Bleed is narrower and more mechanical: the same shopper search, bought several times over by one account, with the resulting data scattered too thin for anyone to use.
Why search term bleed happens
Bleed enters an account through three routes: broad or phrase keywords in one campaign matching searches another campaign was built to own, auto campaigns re-buying search terms that were already harvested into exact match, and branded searches leaking into non-brand campaigns. Three routes, one failure: nobody assigned the search term an owner.
None of these are misconfigurations. Discovery is supposed to wander: auto and broad match exist to find searches you did not know to bid on, and a research campaign doing its job will eventually stumble onto territory your exact campaigns already hold. Bleed starts when the stumbling is never corrected. The most reliable producer is the half-finished harvest: a seller promotes a converting term into exact match, skips the negative exact at the source, and from that day on pays discovery bids to re-find traffic they already own.
| The route | How it starts | What it looks like in the report | The fix |
|---|---|---|---|
| Broad or phrase poaching | A research keyword matches a search your exact campaign already owns | The same search term under different keywords in different campaigns | Negative exact on the owned term in the research campaign |
| Auto re-buying harvested terms | A term gets promoted to exact match but never negated at the source | The auto campaign keeps logging the term at discovery bids | Close the loop: negative exact in the campaign that found it |
| Brand leaking into non-brand | Branded searches match through broad or phrase in non-brand campaigns | Non-brand ACOS looks better than it really is | Negative exact your brand terms in every non-brand campaign |
What search term bleed actually costs
The cost of bleed is statistical before it is financial. A search term needs a minimum number of clicks before any verdict on it (bid up, bid down, harvest, negate) is honest. Bleed divides those clicks across campaigns, so every row sits under the evidence bar while the account as a whole keeps paying.
Run the arithmetic on the dataset I grade across these guides: $36,303 in spend, $111,058 in ad sales, 4,140 orders from 45,672 clicks, which computes to a 9.1% conversion rate, a $0.79 average CPC, and a 32.7% ACOS. At 9.1%, an average order takes roughly eleven clicks, so eleven clicks is what a fair trial costs: about $8.77 of spend, dividing spend by orders. Now split one term's eleven clicks across three campaigns. No row reaches four clicks. Every fragment reads as too little data, forever, while the account has already paid full price for a verdict it never receives. If the term converts, the damage inverts: my harvest bar is two orders, roughly twenty-two clicks at this conversion rate, and no single campaign will ever accumulate them. Bleed multiplies the price of every verdict by the number of campaigns buying the term.
That is the mechanism behind the inflated effective ACOS. The term keeps being bought at whatever the match types happen to charge, in several places at once, and the repricing that would pull it to target (revenue per click times target ACOS, the formula from my guide on lowering ACOS without killing sales) never fires, because no row ever justifies it. The internal competition is for your budgets, not the auction: one search drains several daily budgets at once, including discovery budgets that should be out finding terms you do not own yet. And nothing looks wrong. Each row's spend is modest, each campaign's topline is plausible, and the waste never sorts to the top of any report.
Doesn't Amazon say you can't bid against yourself?
Correct, and the nuance matters. Amazon's auction is account-based: when several of your campaigns are eligible for the same search, only your highest bid enters the auction against other advertisers. Search term bleed does not inflate your CPC through self-competition. Its damage is split data and misallocated budget, which is exactly why it hides so well.
It is also why some operators leave overlap in place on purpose, running a term in exact and auto at once to take two placements on one results page. Deliberate, monitored overlap can work; I have run it. But what I find in the accounts I audit is not a decision made per term. It is an accident nobody noticed, with all of the data cost and none of the intent.
How to spot search term bleed in your own account
The search term report hides bleed by default, because it arrives grouped by campaign: a duplicated term shows up as several unremarkable rows in different sections of the file. Sort the whole report by customer search term instead of by campaign and the duplicates line up next to each other, ready to count.
That one re-sort is most of the diagnostic. Here is the full loop I run:
- Pull the 60-day search term report. Amazon Ads > Bulk Operations > Download, with Sponsored Products Search Term Data ticked. The full walkthrough is in my search term report guide.
- Sort the whole file by customer search term, not by campaign. Or pivot on the term and sum clicks, spend, and orders per term.
- Flag terms with real spend under two or more campaigns. Note which keyword and match type caught the search in each campaign; that column tells you which route the leak took.
- Assign one owner per flagged term. The exact match campaign, holding the term as an exact keyword with its own bid.
- Negate the term everywhere else. Negative exact, campaign level, in every other campaign that bought it.
Two free tools shortcut the manual work. The N-Gram Analyzer decomposes every search term into word-level patterns and aggregates the metrics; the first cleaning step in my n-gram methodology is merging duplicate search terms across campaigns into one row, because fragmented rows lie. And the Audit Dashboard runs a dedicated search term bleed panel, one of the nine checks it performs on a bulk file. If you want to see the read before running your own data, the demo renders it on a real anonymized account.
The fix is ownership: one exact match owner, negatives everywhere else
The fix for search term bleed is ownership. Exactly one campaign owns the term, holding it as an exact match keyword with its own bid, and every other campaign that bought the search gets a negative exact for it. This is the back half of the harvest-then-negate loop, and it is the half most sellers skip.
The owner should be the performance campaign, because that is what the funnel is for: auto and broad exist to find terms, exact exists to run them. If the term converts and no exact keyword exists yet, harvest it first (two or more orders, same-SKU sales, a relevancy check against the live search results) and price the bid at revenue per click times target ACOS. If the term has not earned a harvest, it has not earned three campaigns' budgets either; one campaign is plenty while it finishes its trial.
The negative side takes minutes. Add the term as a negative exact, at campaign level, in every non-owner campaign that bought it. Exact, not phrase: negative phrase would also block every longer variation containing the term, which is precisely the long-tail your discovery campaigns should keep finding. My negative keywords guide covers the match type logic in full, and the free Negative Keyword Finder builds the upload file from your own report. The same ownership rule applied at the tactic level is why brand terms get negated out of every non-brand campaign: without that separation, your non-brand data flatters itself.
What ownership buys you is compounding. The term's whole record lands in one row, so bids get priced on real statistics, losers trip the negation threshold on schedule, and discovery budgets go back to discovering. Re-check for new duplicates monthly while campaigns are changing, quarterly once the account is stable.
A real account where bleed was part of the diagnosis
Search term bleed was one of three named structural problems in the health and personal care account on my results page: bleed across campaigns splitting its conversion data, wasted spend on non-converting keywords, and a bid architecture that rewarded impressions over conversions. The rebuild took ACOS from 39.66% to 27.02% in 90 days.
I will not pretend to isolate bleed's exact share of that 12.64-point drop; the three problems were fixed together, and that is the honest version. But the signature of the fix is worth reading: the account added $4,735 a month in profit with no budget increase, which only happens when the same budget moves from clicks that produce nothing to clicks that convert. Consolidating each term's data under one owner is a large part of what makes that move possible, because you cannot reprice a term you can only see in fragments. The full rebuild, lever by lever, is in my guide on lowering ACOS without killing sales.
Frequently asked questions
What does search term bleed mean in Amazon PPC?
Search term bleed is when multiple campaigns in the same Amazon account buy the same customer search term, usually because broad or phrase match keywords in one campaign match searches another campaign was built to own. The term's conversion data splits across campaigns, so no single campaign has the statistics to optimize it, and its effective ACOS inflates.
Is search term bleed the same as keyword cannibalization?
They overlap but are not identical. Keyword cannibalization usually describes ads eating traffic a listing already earns organically, or two of your products competing for one niche. Search term bleed is narrower and more mechanical: multiple campaigns buying the same customer search term, fragmenting its conversion data. Amazon's auction dedupes your bids, so the damage is statistical, not a bidding war with yourself.
Is it bad to have the same keyword in multiple campaigns on Amazon?
Sometimes it is deliberate and fine. Amazon's auction only enters your highest bid, so duplicated keywords do not bid against each other, and some operators intentionally run a term in exact and auto at once for two placements. It becomes bleed when the duplication is accidental, nobody owns the term, and no single campaign accumulates the data to price, harvest, or negate it honestly.
How do I find duplicate search terms across campaigns?
Download the 60-day search term report from the bulk file, then sort or pivot the whole file by customer search term instead of by campaign; duplicates line up immediately. My free N-Gram Analyzer aggregates the overlapping patterns the same way, and the free Audit Dashboard runs a dedicated search term bleed panel, one of its nine checks, on any bulk file.
Does search term bleed increase my CPC?
Not through the auction. Amazon's auction is account-based: when several of your campaigns qualify for one search, only your highest bid competes against other advertisers. The real costs are split conversion data that never reaches decision thresholds, discovery budgets re-buying terms you already own at exact, and losing terms that never trip a negation verdict because no single row looks bad enough.
Should I use negative exact or negative phrase to fix search term bleed?
Negative exact, almost always. You are blocking one specific search term that another campaign now owns, and negative exact removes only that query. Negative phrase would also block every longer variation containing the term, which is exactly the long-tail traffic your broad and auto campaigns should keep discovering. Reserve negative phrase for irrelevant word patterns, not owned terms.
See whether your campaigns are buying the same searches
The free Account Health Snapshot reads your bulk file and grades nine metrics in about 60 seconds, parsed in your browser, no email, no account; the full Audit Dashboard includes the search term bleed panel itself. And if the report turns up duplicates and the ownership calls are not obvious (they often are not: harvest or negate, which campaign, what bid), the free 30-minute diagnosis call is where I make them with you, term by term.