What is Keyword match type?

A keyword is not a list of searches. It is a statement of what an advertiser wants to appear for, and the match type says how loosely that statement may be interpreted.

Three types exist on Google Ads and Microsoft Advertising, in descending order of latitude.

Broad reaches searches the platform judges related to the keyword, including searches that share no words with it.

Phrase requires the meaning of the keyword to be contained in the search, with additional words allowed around it.

Exact requires the search to mean the same thing as the keyword.

The critical modern fact is in each of those descriptions: all three match on interpreted meaning rather than on literal text. That is a substantial change from how the types were originally defined, and a great deal of advice still in circulation assumes the older behaviour.

Exact match once meant the exact words, in that order, and nothing else. Then it accepted close variants — plurals, misspellings, abbreviations, function words. Then it moved to same-meaning matching, which can include searches with different words entirely.

Phrase match once required the keyword's words to appear in sequence. It now requires the meaning to be included.

Broad match was once a fairly literal expansion around the keyword's words. It is now a semantic reach that draws on the whole context available to the platform, including signals beyond the query itself.

The practical consequence is that an account built on the older understanding is not getting what it thinks it is. Exact-match keywords are matching queries their owner did not anticipate, phrase-match keywords are reaching further than expected, and the assumption that tight match types provide tight control is no longer safe.

Because the types are interpreted rather than literal, the only way to know what a keyword actually reached is the report of the searches that triggered it.

This is the single most important habit in search advertising and the most commonly neglected. A match type is a request; the search-term report is the outcome. An account that reviews keywords and never reviews search terms is managing its intentions rather than its spend.

The report is also where negatives come from. Every irrelevant search that cost money is a negative waiting to be added, and the exercise compounds: a fortnight of disciplined negative work narrows what broad and phrase keywords can reach far more reliably than switching match types does.

Its limitation is coverage. Platforms do not report every search term, particularly those with low volume or privacy implications, so a portion of spend is attributed to searches nobody can see. That portion is a reason to watch the aggregate carefully, not a reason to skip the report.

The relationship between match type and bid strategy is the part most often missed, and it changes what a sensible structure looks like.

Under manual bidding, a broad keyword is a genuine risk. One bid applies to every search it reaches, including the poor ones, and the advertiser has no way to bid less for a bad match than for a good one. Tight match types were the defence, and the defence was necessary.

Under a conversion-based automated strategy, the calculation changes. The bidder evaluates each auction separately and declines the ones it does not expect to convert, which means a broad keyword becomes a way of letting the model find demand rather than an invitation to waste. It is not free — the model learns partly by spending — but the risk profile is very different.

This is why advice about match types dated from the manual era translates badly. The right structure depends on which bidding arrangement is in place, and an account running automated bidding on a tightly exact-matched keyword set may be constraining itself for a reason that no longer applies.

A few consequences worth building around.

Do not duplicate the same keyword across match types in separate groups. The old practice of running exact, phrase and broad versions of everything in parallel structures made sense when the types were literal and bids were manual. With semantic matching and automated bidding it fragments conversion data across entities that each need volume, and the platform arbitrates between them anyway.

Keep groups tight enough for the ad to speak to the queries. Relevance is part of rank, and a group of thirty loosely related terms served by one generic ad reports poor relevance whatever the match type.

Use negatives as the primary control, since they are the only instrument that genuinely refuses.

Check the search terms on a schedule, not when something looks wrong. The drift is gradual and the report is the only place it is visible.

A search that matches more than one keyword in an account does not trigger all of them. The platform picks one, and the rules are worth knowing because they explain a good deal of otherwise confusing reporting.

Where a search matches an exact-match keyword identically, that keyword is generally preferred over looser matches elsewhere in the account. Beyond that, the selection leans on which keyword is most relevant to the search and on expected performance rather than purely on match type or bid.

Two practical consequences. An account cannot reliably force a search into a chosen ad group by adding the keyword there; it can only make that keyword the most plausible candidate. And a keyword showing very little volume is frequently not losing auctions — it is losing the internal selection to another keyword in the same account.

The instrument that does force the outcome is a negative applied to the ad group that should not receive the search, which is another case of subtraction being the reliable control.

For an account with conversion volume and an automated bid strategy: a core of exact-match terms whose value is known, a layer of phrase or broad for discovery, one ad group per tight theme, aggressive negatives maintained from the search-term report, and no parallel duplication of the same keyword across types.

For an account bidding manually or short of conversion data: tighter match types, because the bidder cannot decline a poor match and the advertiser has to do the declining in advance.

The difference between those two recommendations is the whole of the modern match-type argument, and an account that knows which situation it is in can stop having it.