Broad match does not just pull in junk queries — it also quietly steals the good queries your own tighter keywords should have won. When a broad-match ad group and an exact-match ad group are both eligible for the same search, the auction picks one on Ad Rank, not on which is cheaper for you, and the broad group frequently takes a query the exact group would have converted at a better cost. One documented account analysis found broad match cannibalized 35% of budget from higher-performing phrase-match keywords (verdemedia on 2026 keyword strategy). Negative fencing is the fix: use negatives to route each query to its most efficient ad group instead of letting the loosest one grab it.
This is negatives used for a different purpose than blocking waste. In an audit you add negatives to stop irrelevant queries; in fencing you add them to move relevant queries between your own ad groups. The mechanism is the same — a negative keyword makes an ad group ineligible for a term — but the goal is structural: each query should serve on the tightest match type that can match it, because that is almost always the cheapest and best-controlled place for it to run.
How match types cannibalize each other
The core problem is overlap. Broad match can serve on nearly any query related to its keyword, phrase match on any query containing the keyword’s meaning, and exact match on the query and its close variants. Those eligibility sets overlap heavily, so a single high-value query like enterprise crm software can be matchable by a broad keyword, a phrase keyword, and an exact keyword you hold in three different ad groups at once. Google runs one auction and serves one of them. It does not prefer your exact-match keyword just because it is more specific.
Which one wins depends on Ad Rank, and the loose keyword often has enough of it to take the query. The result is that spend you intended to flow through a tightly-managed exact-match ad group — with its own bid, its own ad copy, its clean search terms — leaks into the broad ad group instead, where the bid is coarser and the query sits among a mass of looser matches. Multiply that across your top converting terms and you get the 35%-of-budget cannibalization figure: not wasted on junk, but misrouted into your least efficient structure. Diagnosing it starts in the search terms report, the same place the wasted spend the top-cost view hides lives.
What the fence actually is
A fence is a set of negative keywords that makes a looser ad group ineligible for the queries a tighter ad group should own. The canonical setup separates match types into their own ad groups (or campaigns) and then adds the negative exact variation of each keyword to the broader ad groups, so a query for that exact term is pushed to the exact ad group where it is more efficient (Store Growers on match types). The exact-match ad group is left unfenced; every looser group gets a fence pointing traffic away from it.
Concretely, for a term you want to concentrate in exact match:
- Exact ad group — holds
[enterprise crm software]as a positive keyword. No fence. - Phrase ad group — add the negative exact
[enterprise crm software], so this group serves the broader phrase demand but not that exact query. - Broad ad group — add both the negative exact and the negative phrase for the term, so it captures only the wider discovery queries and never the ones your tighter groups own.
Each query now falls through to the tightest group that can serve it. Discovery queries you have no specific keyword for still land in broad, where they belong; the exact terms you have deliberately built around land in exact, where you control them. The fence does not remove traffic from the account — it decides which of your ad groups gets it.
Fence with exact and phrase negatives, never broad
The match type of the fence is the part people get wrong, and getting it wrong turns a routing tool into a traffic-killer. Fencing works because negative exact blocks only the specific query and negative phrase blocks only queries containing that ordered phrase — narrow, surgical exclusions that move one lane of traffic. A negative broad keyword does the opposite: it can make the ad group ineligible for a huge set of loosely related queries, potentially silencing the broad group you were trying to keep alive for discovery.
So the rule is: fences are built from negative exact and negative phrase only. If you reach for a negative broad to save typing, you are no longer fencing — you are over-negating, and you will throttle the incremental demand the looser group exists to capture. That failure mode, where negatives quietly suppress good traffic, is covered in when negative keywords cost you sales. The behaviour of negative match types — and the plurals-and-close-variants gotchas that decide exactly what a negative blocks — is in negative keyword match types, and it applies to fence negatives exactly as it does to blocking negatives.
Finding the queries to fence
You find fence candidates in the search terms report by turning on the Keyword and Match type columns and looking for queries serving on a loose keyword when you already hold that exact query as an exact-match keyword elsewhere. That mismatch — an exact query landing in a broad or phrase group — is cannibalization you can see. Sort by cost so you fence the highest-spend misrouted terms first; a handful of high-cost queries usually account for most of the leak, and fencing those returns the most budget to your efficient structure per unit of effort.
Do this on a schedule rather than once. Because close-variant matching keeps widening what your broad and phrase keywords can reach, new cannibalized queries appear over time, so re-scan the report every couple of weeks and add fences for the new mismatches. The discipline is the same one behind reading the search terms report for negatives in general: regular short passes beat occasional deep audits, because the account keeps generating new queries between reviews.
When fencing is worth the structure
Negative fencing pays off when you run the same terms across multiple match types and care about controlling cost per query — which describes most accounts past the smallest scale. If you run a single match type, there is nothing to fence; the leak only exists when a looser group can poach a tighter group’s queries. The more you lean on broad match for discovery while keeping exact-match keywords for your proven winners, the more fencing matters, because that is exactly the structure where cannibalization is largest.
Weigh it against the maintenance cost honestly. A fenced account has more negatives to keep straight, and every new exact-match winner you promote needs its fence added to the looser groups or the leak reopens. For accounts with a stable core of high-value terms, that upkeep is trivial next to the budget it reclaims. For a tiny account with a dozen keywords, the simpler move may be to not run overlapping match types at all. Fencing is a structural answer to a structural problem — use it where the structure that creates cannibalization is the structure you actually want.