Finding negative keywords is only half the job. The other half is deciding where each one lives — and most accounts get this wrong by dumping everything into one place. Google Ads gives you four levels to house a negative: the account level, shared negative keyword lists, the campaign level, and the ad group. Each has a different reach, a different cap, and a different job. Put a negative at the wrong level and you either waste a scarce slot, block a query one campaign wanted, or lose track of why traffic dried up.
This post is about the architecture, not the hunt. If you still need to find candidates, start with the search terms report walkthrough and the n-gram analysis method. Once you have a list of terms to block, this is the map for where they belong: the three main levels, the hard caps that force the structure, what belongs at each, and how the whole thing behaves under Performance Max and AI Max.
The three places a negative can live
A negative keyword sits at one of three practical scopes, from widest to narrowest. Account-level negatives apply automatically to every eligible campaign in the account — Search, Shopping, Performance Max, Smart, App and Local — with no per-campaign setup. Shared negative keyword lists are reusable themed sets you create once and attach to whichever campaigns should inherit them, so a single list can guard many campaigns at once. Campaign-level negatives live inside one campaign and apply only there. Beneath those sits the ad group level, the finest scope, used less to block waste than to steer a query toward the right ad group when two could serve it.
The mistake is treating these as alternatives — picking one and using it for everything. They are not alternatives; they are a hierarchy, and a well-run account uses all of them at once. The account level carries the universal junk, shared lists carry the reusable themes, campaign negatives carry the exclusions that are only true for one campaign, and ad-group negatives handle the rare sculpting case. Getting a negative to the right level is what keeps the account both protected and legible: you can look at any query that was blocked and know immediately which layer stopped it and why.
The caps that force the architecture
The reason structure matters at all is that the levels have very different ceilings, and the widest-reaching level has the smallest one. Account-level negatives are capped at 1,000 keywords. Each shared negative keyword list holds up to 5,000 keywords, you can keep up to 20 lists per account, and a single list can be applied to up to 1,000 campaigns at once. Campaign-level negatives allow up to 10,000 keywords per Search campaign. Read those numbers together and the design writes itself: the account level is deliberately scarce, so it cannot be your catch-all.
That scarcity is a feature. Because you only get 1,000 account-level slots, you are forced to reserve them for negatives that are genuinely universal — the terms you never want to pay for in any campaign. The bulk of your themed exclusions, which can run to thousands of terms, belong in shared lists where the 20×5,000 headroom lives, and the one-off campaign-specific blocks belong in the campaign itself. As Google’s own documentation on negative keyword lists lays out, shared lists exist precisely so you can categorise negatives by theme and apply them across campaigns without retyping — which is the whole point of the middle tier.
What belongs at the account level
The account level is for negatives that are true everywhere, forever. Brand-safety terms you never want your ads near, entire off-topic industries, adult or violent modifiers, and the evergreen junk — free, torrent, crack, salary — that no campaign in your account would ever want to pay for. These are set-and-forget: because they apply automatically to every eligible campaign including Performance Max, one entry covers the entire account, which is exactly the leverage you want from your scarcest 1,000 slots.
The discipline here is restraint. Every term you add at the account level is a term that can never be un-blocked for a single campaign without removing it globally, so anything that might be wanted somewhere does not belong here. If a term is junk for your lead-gen campaign but valuable for a content campaign, it is not a universal — push it down a level. Treat the account list like a constitution: short, stable, and only amended for things that are true account-wide. A bloated account-level list is one of the quiet ways good traffic gets accidentally throttled, because a broad universal negative can catch queries a specific campaign would have converted.
What belongs in shared lists
Shared lists are the workhorse tier, and the one most accounts underuse. They exist for themed, reusable exclusions — the sets of terms that apply to some but not all campaigns and recur across them. A “job seekers” list (jobs, careers, hiring, salary), a “free / cheap / DIY” list, a “competitor brands” list, a “wrong intent” list — each is built once and attached to the campaigns that share that risk. When you launch a new campaign, you inherit years of accumulated protection by ticking a few boxes instead of rebuilding the list from scratch.
The key to shared lists is naming and single-theme discipline. Give each list one clear job and a descriptive name, so that when a query is wrongly blocked you know instantly which list to open, and so that a colleague inheriting the account can read the structure without guessing. Resist the urge to create one giant “all negatives” list — that destroys the reusability, because you can no longer attach the “job seekers” theme to a campaign without also forcing every other exclusion onto it. Small, themed, well-named lists are what make the middle tier auditable. This is also where your n-gram findings usually land: a recurring wasteful word-pattern across many campaigns is a shared-list candidate, not an account-level one.
What belongs at the campaign and ad group level
Campaign-level negatives are for exclusions that are only true for one campaign. A term that is junk for your generic campaign but is the exact product your brand campaign sells, a competitor you block in prospecting but bid on in a conquesting campaign, a seasonal exclusion that only applies to one push — these are campaign-specific by nature and would cause damage if applied globally. With up to 10,000 keywords per Search campaign, there is ample room for detailed, surgical blocking that reflects that one campaign’s intent without touching anything else.
Ad-group negatives are the finest tool and the one used least for waste. Their main job is sculpting: when the same query could match two ad groups, an ad-group negative in one forces it into the other, so your tightly-themed ad copy and landing page win the impression rather than a generic one. Used this way they improve relevance rather than block spend. Used carelessly — scattering the same negative across ad groups instead of lifting it to the campaign or a shared list — they create an unmaintainable mess where nobody can tell why a query is or is not eligible. The rule of thumb: push a negative as high as it can go while still being true, and no higher.
How the levels behave under Performance Max and AI Max
Performance Max inherits your account-level negatives automatically, which makes the account level the single simplest lever for keeping PMax out of obviously wrong queries. PMax also accepts its own campaign-level negative keyword list and can have shared lists applied to it, so the same three-tier logic holds. The one caveat is scope: negatives on PMax govern its Search and Shopping query eligibility, not its YouTube, Display, Gmail or Discover placements — so a negative keyword is a query filter there, not a universal channel exclusion. The mechanics of adding them are covered in Performance Max negative keywords.
AI Max complicates the picture because its expanded query matching pulls in more queries than your keywords alone would, which raises the value of a clean negative architecture — but it can also reinterpret or override negatives in ways that surprise people. Before you assume a negative at any level is doing its job under AI Max, read when AI Max overrides your negative keywords, because a well-structured list still needs verifying against what actually served. Structure makes the account legible; it does not exempt you from checking the search terms report to confirm the blocks held.
Putting it together: a placement decision
When a new negative candidate comes off your search terms review, run it through one question: how widely is this true? If it is junk for the entire account, it goes account-level — sparingly, because you only have 1,000 slots. If it is a themed exclusion that recurs across several campaigns, it goes in the relevant shared list, created once and reused. If it is only true for one campaign, it goes in that campaign. If it exists to steer a query between two ad groups, it goes at the ad group. Push each negative as high as it can go while remaining true everywhere it applies, and no higher.
Maintained this way, the account stays both protected and readable: account-level for universals, shared lists for reusable themes, campaign-level for the surgical, ad-group for sculpting. The payoff is not just tidiness — it is that when traffic behaves oddly, a legible structure tells you exactly which layer to inspect, instead of leaving you scrolling one undifferentiated list of thousands of terms wondering which one blocked the query that should have converted. Good negative-keyword hygiene is as much about placement as it is about the list itself.