What is Phrase match?
Phrase match makes a keyword eligible for searches that contain its meaning, with additional words allowed around it. It no longer requires the keyword's literal words in order, which is the change that makes its name misleading.
A containment test
Phrase match makes a keyword eligible for searches that include the keyword's meaning, with other words permitted around it.
The test is containment. The search has to be about the thing the keyword names; it may also be about other things, and it may express the meaning in different words.
That last clause is the modern part and it is what the name no longer conveys. Phrase match once required the keyword's literal words, in order, with additions allowed only before and after. It now matches on interpreted meaning, and a search containing none of the keyword's words can satisfy the containment test if the platform judges the meaning present.
Where it sits
Between the other two, and the position is more useful than it sounds.
Exact requires the search to mean the same thing as the keyword. The set is narrow and the advertiser largely knows what is in it.
Phrase requires the keyword's meaning to be part of the search. The set is larger, and it is bounded — every member has to contain the meaning.
Broad requires only relatedness. The set is largest and the bound is the platform's judgement rather than a test the advertiser can reason about.
The bounded-versus-unbounded distinction is the one that matters operationally. Surprises from phrase match tend to be adjacent: the keyword's meaning plus something unexpected around it. Surprises from broad match can be genuinely alien, because relatedness is a much weaker constraint than containment.
Why word order still sometimes matters
The relaxation of strict ordering is not a removal of ordering entirely, because order sometimes carries meaning.
A keyword naming a journey from one city to another does not match a search naming the reverse journey, since the two mean different things and the containment test fails. The same applies to any construction where reversing the terms inverts the sense.
Where order carries no meaning — an adjective before or after a noun, two attributes in either sequence — the variants are treated as equivalent.
This is a sensible behaviour and it is one more reason the literal reading of the name misleads. The setting is doing semantic work and using a syntactic label.
When it earns its place
Accounts that cannot support broad match. An account short of conversion volume, or bidding manually, has no per-auction filter. Broad match without a filter is expensive; phrase provides a structural limit that does not depend on the bidding being clever.
Terms where a specific meaning is load-bearing. Where traffic is only worth having if a particular concept is present — a product category, a service, a qualifier like "commercial" or "wholesale" — the containment test enforces it more efficiently than a negative list trying to remove everything that lacks it.
A middle layer during discovery. Some accounts run exact for known value, phrase for controlled expansion, and broad only where the bidding is strong. That is a defensible structure and it is not the old practice of duplicating every keyword across all three types, which fragmented data for no benefit.
When it is the wrong choice
As a substitute for negatives. Containment does not mean relevance. A search can contain the keyword's meaning and still be worthless — informational, job-seeking, competitor research, or a person looking for a free version. Negatives remain the instrument that refuses.
Under an account with strong automated bidding and good conversion volume. Here broad match plus a filtering bidder generally finds more without costing more, and phrase is leaving reach unused.
Duplicated alongside exact and broad versions of the same keyword. The platform arbitrates between an account's own candidates anyway, and the parallel structure splits conversion data across entities that each need volume.
What to watch
The search-term report, on a schedule. It is the only view of what the containment test actually admitted, and it is where the negatives come from.
The share of spend on terms you would not have written. Some is discovery and worth having; a rising share is drift, and it usually means either the keyword is more ambiguous than it looked or the ad group's context has broadened.
The relationship between phrase keywords and exact ones in the same account. A phrase keyword producing consistently valuable searches is telling you which exact keywords to add, and harvesting them moves that traffic into a structure where the ad can speak to it directly.
And the ad group's tightness, because context influences matching. A phrase keyword surrounded by unrelated terms reaches differently from the same keyword in a coherent group, which is one of the quieter reasons two accounts with identical keyword lists get different results.
The modifier that disappeared
Accounts that have been running for a while may still carry the remains of broad match modifier, a fourth type that required certain words to be present while allowing any order around them.
It was retired and folded into phrase match, which absorbed its behaviour. Keywords that used the modifier syntax continue to function as phrase keywords, and the plus signs that marked them are inert.
Two practical consequences. Documentation and account conventions written before the change describe a type that no longer exists, which is worth knowing when inheriting an account or reading older advice. And an account still organised around the distinction between modified broad and phrase is maintaining a structure that has no behavioural basis, which is a candidate for consolidation.
A note on how to decide
The match-type question is answered more reliably by looking at the account than by reasoning about the types.
Ask what is doing the filtering. If a conversion-based bid strategy with real volume is declining poor auctions, the account can afford width and phrase is leaving reach unused. If nothing is filtering per auction, the containment test is doing the work and phrase is the sensible limit.
Ask what the traffic is worth without the core meaning present. If a search that lacks the keyword's concept is worthless, containment earns its place. If the concept is incidental and the value comes from intent expressed some other way, containment is an arbitrary restriction.
And ask who maintains the negatives. Every loosening increases the maintenance, and a setting nobody has time to service will drift.
Phrase match versus Broad match
Phrase match is the bounded one of the two. It admits a search only when the meaning of the keyword is contained in it: extra words are allowed around the phrase, but the search still has to be about the thing the keyword names. Broad match asks a looser question — whether the platform judges the search related — which can be true of searches that share no words with the keyword at all. That difference decides how much of the filtering the advertiser does in advance and how much the bidder does per auction. A phrase keyword is filtered by the containment test before the auction, so the surprises in the search-term report tend to be adjacent rather than alien and a manual bid is survivable. A broad keyword is filtered by the bidder, so it wants a conversion-based bid strategy to decline the poor matches and a maintained negative list to refuse the specific ones; with a manual bid and no negatives it is the most expensive setting in search advertising. Read the pair as a choice about where the control sits rather than as a ranking. Phrase buys predictability and pays for it in the searches it never sees; broad buys reach and pays for it in the maintenance it needs.