Here's a fact that explains most leaking negative keyword lists, and it's rarely stated plainly: adding free as a negative does not block freee, freebie, or in some setups even frees. Negative keywords don't behave like positive ones, and almost everyone assumes they do.
The asymmetry that costs you money
Positive keywords are generous. Bid on running shoes and Google will match misspellings, singular and plural forms, stemmings and — depending on match type — related searches you never typed. That's close variant matching, and it's been expanding for years.
Negative keywords are literal. They match what you wrote and nothing else. No misspellings, no plurals, no stemming, no synonyms.
The rule to internalise Positive keywords expand to catch traffic. Negative keywords do not expand to block it. Every variant you want excluded has to be typed out.
The practical consequence is severe. You add job to stop paying for people looking for employment, and you keep paying for jobs. You add free and keep paying for freebies. You add a competitor's name and keep paying for the version people misspell — which, for most brand names, is a meaningful share of the searches.
So a list that looks thorough can still be leaking on exactly the terms you thought you'd handled. When you add a negative, add its family: plural, common misspellings, the abbreviated form, the hyphenated and unhyphenated versions.
One further quirk worth knowing: negative keyword matching is only evaluated against roughly the first sixteen words of a search query. Long conversational searches — increasingly common — can slip past a negative sitting near the end.
Negative match types work differently too
The second layer of the same asymmetry, and where over-negation usually starts.
| Negative type | Blocks | Risk |
|---|---|---|
Broad cheap shoes |
Any query containing both words, in any order, with anything else | Highest — blocks far more than intended |
Phrase "cheap shoes" |
Queries containing that exact sequence | Moderate — usually the right default |
Exact [cheap shoes] |
Only that query, precisely | Lowest — and leaks the most |
Most guidance says "use broad for maximum coverage." That's how accounts end up blocking traffic they wanted. A broad negative on cheap kills every query containing the word — including someone searching for a cheaper alternative to an expensive competitor, who may be an excellent prospect.
Phrase match is the sensible default for most negatives. Reach for broad only when you're certain every query containing those words is worthless, and for exact when you're surgically removing one specific search.
What changed in Performance Max
If your knowledge of PMax negatives is from 2023, it's out of date in ways that matter — and a surprising amount of currently-published advice hasn't caught up either.
PMax launched with a hundred negative keyword slots for a campaign type serving across search, shopping, display, YouTube, Gmail, Discover and Maps. Advertisers reported substantial shares of PMax spend going to queries they simply couldn't block, and adding campaign-level negatives originally required contacting Google support.
Through 2025 that changed considerably:
- Campaign-level negatives became directly addable in the interface around the turn of 2025, without needing a support request.
- The per-campaign limit went from 100 to 10,000 in March 2025, matching Search campaigns.
- A PMax search terms report arrived the same month, giving visibility into what PMax was actually matching.
- Shared negative list support followed later in 2025, which matters enormously for anyone managing multiple accounts.
You can't negate what you can't see
The structural limit on this entire discipline, and the part most "complete guides" skip.
Google restricted search term reporting in 2020 on privacy grounds, withholding queries searched by too few people. Estimates commonly place the hidden or aggregated portion somewhere around a fifth to a third of queries, grouped under a general heading rather than listed.
Which means some proportion of your spend goes to searches you cannot identify and therefore cannot exclude. No negative keyword strategy is complete, and any guide promising you'll eliminate wasted spend is overselling.
Two things follow. First, the visible terms still carry most of the spend, so the work remains worth doing — this is a reason for realistic expectations, not for giving up. Second, when the search term report can't explain a performance problem, stop interrogating it and look at structural fixes instead: tighter match types, better campaign segmentation, audience signals, or landing page relevance.
It's also one more instance of the broader measurement squeeze running through why attribution keeps getting harder — the platforms increasingly know things about your spend that they don't show you.
Landingi
Drag-and-drop landing page builder with reusable Smart Sections, A/B testing and 400+ templates.
Best for: Marketers and agencies building pixel-perfect landing pages without code.
Where the waste actually lives
Six categories cover the large majority of genuinely wasted spend. Working through these is faster than reading the whole report line by line.
Job seekers. jobs, careers, salary, hiring, vacancy, internship. Reliably the largest single category for B2B advertisers and almost always worth an account-level block.
Free and DIY. free, diy, template, tutorial, how to make. Careful here — if you offer a free tier or trial, blocking free broadly will remove qualified traffic.
Research intent. what is, definition, meaning, examples, wikipedia. Whether these are waste depends on whether you're running an awareness play or a direct-response one, so decide deliberately rather than blocking by reflex.
Wrong product or adjacent category. The account-specific ones. If you sell software and get queries for hardware, or sell services and get queries for products, these are usually the highest-value negatives available and they're invisible on any generic list.
Wrong audience or geography. Student, wholesale, or place names outside your service area. Note that geographic negatives are a blunt instrument compared to location targeting — use targeting first.
Second-hand and cheap. used, second hand, refurbished, cheap, discount. Only if you genuinely don't serve that market.
The pattern in that list: the first three are generic and belong in a shared list applied everywhere. The last three are account-specific and can only come from your own search term report, which is why downloadable "500 negative keywords" lists get you started and never get you finished.
Finding the terms in the first place
The search term report is the only trustworthy source for new negatives, because every other source — a downloaded list, last quarter's spreadsheet, instinct — is a guess about queries your account may never actually serve.
Two habits make it faster. Filter by cost first and read top-down; the expensive irrelevant term matters more than the frequent one. And filter for terms with clicks but zero conversions over a meaningful window, which surfaces the queries that look plausible enough to click and never do anything — usually the most valuable and least obvious negatives available.
It's also worth working the problem from the other end. Negatives fix the symptom; match type and campaign structure fix the cause. If a single ad group is generating most of the junk, the underlying issue is often a keyword that's too broad for the intent you wanted, and tightening it removes a whole class of future negatives. That's the same logic as doing keyword research properly — deciding what you actually want to match, rather than casting wide and correcting afterwards.
Over-negation: the failure nobody warns about
Cutting waste is safe advice that produces unsafe behaviour when taken too far.
Three ways aggressive negation costs more than it saves. Broad negatives block adjacent intent — the cheap example above. Conversion data thins out, and Smart Bidding needs volume to work; starving an algorithm of signal to save a small percentage of spend is frequently a bad trade. And you lose learning, since a query that converts unexpectedly is information you'll never get if you blocked it pre-emptively.
Practical guards: check impression and conversion volume for a week after any large addition; prefer phrase over broad; and when a term is merely low-performing rather than obviously irrelevant, consider bid adjustments or a separate campaign rather than exclusion.
Worth remembering too that a wasted click sometimes isn't the ad's fault. If irrelevant-looking traffic converts, or relevant traffic doesn't, the problem may sit after the click — the domain of landing page design and end-to-end funnel auditing rather than of keyword exclusion.
Structure: where to put them
Three levels, and getting this wrong creates work you'll repeat forever.
Account level — applies everywhere, no exceptions. Reserve for things that are never relevant to your business under any circumstance: job-seeker terms, adult content, obviously unrelated categories. Small list, high confidence.
Shared lists — the workhorse. Build themed lists (job seekers, free/DIY, competitors, wrong-category) and apply them to the campaigns that need them. One edit propagates everywhere, which is the entire point, and it's how you avoid maintaining the same twenty terms in eleven campaigns.
Campaign and ad group level — for genuine exceptions and for sculpting traffic between your own ad groups.
The most common structural mistake is putting everything at campaign level, which turns every future update into eleven separate edits and guarantees the lists drift apart. If you're setting up a new account, build the shared lists before launch — the same discipline that makes a well-structured first campaign pay off later.
Judge it on the right number
One measurement caution. The obvious way to evaluate negative keyword work is money saved, which is both easy to calculate and misleading — blocked spend isn't saved unless it would genuinely have been wasted, and the counterfactual is unknowable.
The more honest read is whether cost per acquisition improved while conversion volume held. If both spend and conversions dropped proportionally, you didn't cut waste; you cut reach. Watching those two numbers together is the difference between an optimisation and a quiet reduction in results, and it belongs in whatever KPI framework you report against.
A note on Amazon and retail media
Half the guidance on this topic is written for Amazon PPC, where the mechanics differ enough to matter.
Amazon offers negative phrase and negative exact only — there's no broad negative. And in auto campaigns, negative targeting is often less useful than simply structuring campaigns so that discovery and harvesting are separated: let auto campaigns find converting terms, move those into manual campaigns as exact-match targets, then negate them in auto so the two don't compete for the same query.
That negate-after-harvest workflow is the core of Amazon negative keyword practice and has no exact equivalent in Google Ads. If you're running both, don't port habits across — the platforms reward different structures. Retail media networks beyond Amazon vary again, each with their own controls, which is worth factoring in as more budget shifts into those environments.
A workflow that survives a busy month
- Before launch: apply your shared lists. Starting from zero means paying to learn things you already knew.
- Week one, daily: new campaigns generate the most junk while the system is learning. Ten minutes a day for the first week catches the worst of it.
- Weekly while spending heavily: sort the search term report by cost, descending. Work top-down and stop when terms stop being obviously wrong — the tail rarely repays the time.
- Fortnightly once stable: new irrelevant terms slow to a trickle as the account matures.
- Quarterly: review the negatives themselves. Lists accumulate terms that made sense once. Blocking something you now sell is a genuinely expensive filing error.
- Always: add the family, not just the term. Plural, misspelling, abbreviation.
Sorting by cost rather than by impressions is the highest-leverage habit in that list. A term with 4,000 impressions and no spend costs you nothing; a term with 40 impressions and significant spend is where the money is going.
On small budgets this discipline matters disproportionately — a few pounds a day lost to job-seeker traffic is a large share of a small account, which is one of several reasons running paid ads on a limited budget demands tighter control than a large one.
The short version
Negative keywords don't match close variants, so every plural, misspelling and abbreviation needs adding explicitly — that asymmetry is why lists that look thorough still leak. Use phrase match as the default rather than broad, because over-negation blocks adjacent intent and starves Smart Bidding of signal. PMax negatives improved substantially through 2025 and a lot of published guidance still describes the old limits. Accept that a fifth to a third of search terms are hidden, so completeness isn't available. Sort by cost, work top-down, add the whole family of each term, and audit your own list quarterly for terms that have outlived their purpose.
Not sure how much of your ad budget is going nowhere?
We audit paid accounts, cut the waste, and rebuild the structure so it stays cut.
Explore Performance Marketing →Frequently asked questions
Do negative keywords match close variants like misspellings and plurals?
No, and this is the single most consequential misunderstanding in negative keyword management. Positive keywords match close variants automatically — misspellings, singular and plural forms, stemmings, and often synonyms. Negative keywords do not. Excluding the word free will not exclude freee or freebie, and excluding job will not exclude jobs. Every variant you want blocked must be added explicitly, which is why lists that look comprehensive still leak spend on obvious junk.
Can you add negative keywords to Performance Max campaigns?
Yes, and the position improved substantially through 2025. Campaign-level negatives began rolling out to Performance Max around the turn of 2025, the per-campaign limit was raised from 100 to 10,000 in March 2025, a Performance Max search terms report arrived the same month, and shared negative list support followed later in the year. Some guides still describe the old situation where only account-level negatives were available, so check the current interface rather than relying on documentation written before these changes.
How many negative keywords can you add in Google Ads?
Search campaigns support up to 10,000 negative keywords per campaign, and Performance Max matched that limit in March 2025. Shared negative keyword lists have historically been capped at 5,000 keywords each with up to 20 lists per account, though some advertisers have reported higher limits since late 2025. Account-level negatives, which apply across every campaign, are more tightly capped. In practice most accounts never approach these ceilings — the constraint is knowing which terms to add, not room to add them.
Why can't you see all the search terms triggering your ads?
Google restricted search term reporting in 2020 on privacy grounds, withholding queries that too few people have searched. Estimates commonly place the aggregated or hidden portion around a fifth to a third of queries, grouped under a general heading rather than listed individually. The practical consequence is that a share of your spend goes to searches you cannot identify and therefore cannot exclude, so no negative keyword strategy can be complete — the visible terms still carry most of the spend and remain worth working through.
Can you add too many negative keywords?
Yes, and over-negation is a real and under-discussed failure. Broad match negatives block any query containing all the specified words in any order, so a negative like cheap can eliminate legitimate commercial searches, and an overly aggressive list can starve Smart Bidding of the conversion data it needs to work. The safest habit is to use phrase or exact match for negatives unless you specifically want the wider block, and to check impression volume after any large addition rather than assuming nothing valuable was lost.
