Negative Keyword Match Types Don't Work Like Positives

The three match types, and the punctuation that sets them

You set the match type with punctuation: no punctuation is broad, "quotes" are phrase, [brackets] are exact. Same syntax as positive keywords, different behaviour.

Take the negative plumber salary for a plumbing lead-gen account:

For the same string, broad is the widest and exact the narrowest. Broad ignores word order and allows extra words. Phrase requires the words adjacent and in order, but allows extra words before and after. Exact allows nothing extra.

One practical detail that catches people out: when you paste a list of negatives into the Google Ads UI or into Editor without punctuation, every line is added as broad. If you built your list in a spreadsheet and stripped the quotes somewhere along the way, you have just applied the widest match type to every row.

And for a single word, broad and phrase do the same thing. free and "free" both block any search containing the token "free". The match type only starts to matter from two words up. That sounds obvious once written down, but it means half the arguments about broad versus phrase are arguments about nothing.

A broad negative does not expand like a broad keyword

This is the mistake that costs the most, and it costs money silently.

A broad keyword is a request for Google to interpret. Bid on dog crate in broad match and you can be served on "puppy cage", "travel kennel for a labrador", "where to buy a crate for my dog". Google matches meaning, not words. That is the whole point of the match type.

A broad negative does not interpret anything. It matches the literal words. dog crate diy as a broad negative blocks searches that contain the words "dog", "crate" and "diy" in some order. It does not block "build your own dog crate". It does not block "dog crate plans". It does not block "homemade puppy cage". The word "diy" is not there, so the negative does not fire.

So the words "broad" and "phrase" mean almost opposite things on the two sides of the account. On the positive side, broad is wide in meaning. On the negative side, broad is wide only in word order. Nobody blocks a theme with a broad negative. They block a set of words.

The cost is not the queries that get through. It is what you stop doing. You add plumber salary as a broad negative, you write "job seekers — done" on your list, and you stop reading that part of the search terms report. Meanwhile the account keeps paying for "plumber wages uk", "plumbing apprentice pay", "is plumbing well paid", "plumber hourly rate" — and "plumber hourly rate" is genuinely ambiguous, some of those people are hiring one.

There is a structural asymmetry here worth sitting with. Your positive keywords expand on their own: Google adds new interpretations, new synonyms, new query shapes, without you doing anything. Your negatives only grow when you type. The set of searches you can be charged for widens by itself. The set you have blocked does not. Every week you do not look at the search terms report, the gap gets wider in one direction only.

Single-word broad negatives are the bluntest tool in the account

Broad negatives are narrower than people expect in meaning and wider than people expect in reach. The second half bites when the negative is one or two words.

free as a broad negative on a pet supplies account blocks "dog crate free shipping". Someone searching for free shipping is holding a credit card.

cheap blocks "cheap 42 inch dog crate" — which for a value retailer is the best query in the account.

near me blocks "emergency plumber near me now" along with "plumber training courses near me".

used blocks "used dog crate" and also "replacement tray for used dog crate", which is a spare parts sale.

None of that is exotic. What makes it dangerous is the feedback loop, and this is the part most articles on match types skip: Google gives you no report of what a negative keyword blocked. A suppressed search does not appear in the search terms report with zero impressions. It does not appear at all. There is no "blocked by negative" column, no negative keyword performance view, no history of what a term would have earned. The evidence of your mistake is deleted by the mistake.

So a bad single-word broad negative looks like nothing. No error, no warning, no line item. Impressions drift down, conversions drift down, and three weeks later somebody blames the algorithm or the season. A bad bid or a bad budget shows up in the interface. A bad negative shows up as absence.

That is the real reason to be careful with broad, and it is a stronger reason than "it might be too wide". Reserve single-word broad negatives for tokens that can never be commercial in your account — "salary", "vacancies", "wikipedia", "lyrics", "meaning", "pdf" — and before you add even those, filter the search terms report for the token and check that nothing with the word in it has ever converted.

Negatives ignore close variants, so plurals and typos get through

Google matches positive keywords to close variants: plurals, misspellings, stems, accents, abbreviations. Negative keywords do not get close variant matching. Google documents this. The negative you type is the negative you get.

Concretely, [plumber salary] as an exact negative does not block:

• "plumbers salary" (plural)
• "plumber salarys" (misspelling)
• "plumer salary" (misspelling)
• "plumber salery" (misspelling)

And "dog crate" as a phrase negative does not block "dog crates", because "crates" is a different string.

There is no match type that solves this. Broad fixes word order. Phrase fixes adjacency. Nothing fixes spelling. If you want the plural blocked, you type the plural. If you want the two most common typos blocked, you type them too.

This is why manual negative work never reaches an end state. The tail is not a fixed list of 300 bad queries you can clear in an afternoon. It is a generator. Every week produces new spellings, new word orders, new modifiers, and each one needs its own row or a phrase negative wide enough to catch it. If you have ever wondered why the same waste keeps reappearing after you cleaned it, this is most of the answer — the rest is in how to find wasted ad spend in Google Ads.

Negatives always beat your positive keywords, including exact ones

A negative keyword is evaluated before anything else. If the search matches a negative, the auction is over. It does not matter that you have that exact query as an exact match keyword with a high bid and a good Quality Score. Negative wins.

The usual way this shows up:

• A shared negative list built for the main campaign gets applied to a new campaign that legitimately sells the blocked thing. Classic case: a list containing used applied to the campaign that sells refurbished stock.
• A broad negative repair sits at campaign level while an ad group is built on the keyword [boiler repair]. The ad group serves nothing. It looks like a search volume problem, and gets "fixed" by broadening the keyword, which changes nothing.
• Someone adds "quote" as a phrase negative to filter out "insurance quote" style traffic, and takes out "plumber quote online" with it.

When impressions collapse with no bid, budget or keyword change, check negatives at every level before you check anything else. There are four places to look: ad group negatives, campaign negatives, shared negative lists applied to the campaign, and account-level negative keywords. Account-level negatives are the ones people forget, because they were added later and live somewhere different in the interface.

How to choose the match type, per term

Default to phrase for a theme, exact for a specific query you want to stop paying for today, broad only for a single token that can never be commercial. Then check the actual wording in the report before you commit.

A workable order of questions:

Is the whole search irrelevant, or just one word in it?

"plumber salary" is irrelevant because of "salary". Every search containing "salary" is irrelevant for a plumbing lead-gen account. That is a token-level problem: one negative, salary, handles the entire family — salary, salary uk, average salary, salary vs wage — in one row. This is the case where a one-word negative earns its keep.

"dog crate rental" is different. "rental" might be irrelevant today, but "crate rental for moving house" could be a product line next year, and "rental property dog crate" is a buyer. Block the specific query, not the token.

Does the word order matter?

If the offending words only ever appear together in one order, phrase is enough and safer, because it cannot fire on an odd combination you did not picture. "how to make" as a phrase negative is predictable. how to make as a broad negative also blocks "make an appointment how to" — unlikely, but you are giving up the ability to predict.

Where the order really does vary, broad is the right answer: plumber salary broad catches both "plumber salary uk" and "salary of a plumber", which would need two phrase negatives.

Write from the literal words, not from the intent

This is the habit that separates a list that works from a list that feels like it works. Open the search terms report, read the actual strings you paid for, and build negatives from those strings. Do not sit in a meeting inventing themes. Your intuition speaks English; negative keywords speak substrings.

One more caution on broad negatives across match types

The same broad negative behaves differently depending on what it collides with. Against a tightly built exact-match ad group it is almost inert. Against a broad-match keyword, or against Shopping and Performance Max where there are no keywords at all and the query set is entirely Google's choice, a broad negative is the main thing shaping which searches you can appear on. The reach of free in a Shopping campaign is not something you can work out on paper. Add it, then watch impressions for a week.

Match types in Performance Max, where the 10,000 cap changes the maths

Performance Max accepts up to 10,000 campaign-level negative keywords, in all three match types. It does not accept shared negative keyword lists. Standard Shopping accepts both negatives and shared lists, exactly like Search.

Two consequences for match type choice.

First, every Performance Max campaign carries its own list, so a rule you want across twelve campaigns is twelve uploads and twelve maintenance jobs. Match type discipline matters more when the same decision has to be replicated.

Second, the 10,000 rows are a finite resource, and match types spend it at very different rates. One phrase negative "salary" covers a family that would take dozens of exact rows. If you block query by query in exact match, a busy account will grind toward the ceiling over a few years. If you block by theme in phrase match, you stay well clear of it — and you accept a wider blast radius in the one campaign type where you can least inspect the damage, because Performance Max search term reporting is coarser than Search. That trade-off, and what to do when you hit the wall, is in negative keywords in Performance Max and the 10,000 cap.

Worth stating plainly: search terms exist in campaign_search_term_view for Search, Shopping and Performance Max only. Display, Video and App report no search terms. Adding negative keywords with any match type to those campaign types is not something you can measure, because you never see the queries in the first place.

What match type PPCleaner uses, and what that costs you

PPCleaner adds exact negatives. It blocks the specific search term it saw in the report, word for word, and nothing else. Collateral damage is zero by construction: an exact negative cannot take out a query you have never seen, because it only matches one string.

That is a deliberate choice, and it has a price. Here it is.

You pay once for each variant. If the report contained "plumber salary" but not "plumber salary uk", the second one is not blocked. It will get one click, appear in tomorrow night's report, and be blocked then. Across a long tail that means the account pays a single click's worth of tuition per variant rather than one click per variant per month, forever. It is not zero. A hand-written phrase negative on "salary" would have caught all of them in advance — and would also have caught "plumber salary vs electrician salary near me", which nobody wanted, and "is a plumber salary enough for a mortgage", which nobody wanted either, so in that particular case phrase is clearly better and you should add it yourself.

Exact rows eat the Performance Max cap faster. If a Performance Max campaign is approaching 10,000 negatives, an exact-per-term approach is the wrong shape for the last stretch and needs pruning or consolidating into phrase negatives by hand.

It does not audit the negatives you already added. If a broad negative you applied two years ago is strangling an ad group, that is your list, and PPCleaner will not touch it. It never touches budgets, bids, ads or keywords either. Diagnosing an over-broad negative is still a human job with four levels to check.

The two things it does give you back are the ones Google does not. It never blocks a term that converted in the last 90 days, so the worst case of an automated block — killing a proven query — is off the table. And every block can be undone in one click for 90 days, which matters precisely because of the problem described earlier: Google will not tell you what a negative suppressed, so the only real safety net is being able to put it back.