Google Ads phrase match: close variants took 71% of spend at a 45% CPL premium

We took every search term that triggered an ad in a ten-day window (22,717 rows across the enabled search campaigns of a mid-sized account), and asked one question of each: does the query contain every word of the keyword it matched?

That is a deliberately literal test. It doesn't ask whether the query was relevant, or whether it converted, or whether a human would have approved it. It asks only whether the advertiser wrote those words. If they didn't, the match was Google's decision, not theirs.

Seventy-one percent of the money went to queries the advertiser never wrote, and those queries cost 45% more per lead.

Where the close-variant spend went

The ten keywords responsible for the most non-matching spend:

Two of the ten spent over a thousand dollars each on close variants that produced nothing. solar panel system alone spent $8,446 on queries that weren't "solar panel system", at $655 per lead, more than double the account's non-brand average.

This is not an argument that close variants never convert. They produced 72 conversions in ten days, more than the literal matches did. The argument is that each one cost 45% more, the advertiser didn't choose to buy them, and a meaningful fraction of them were pure waste that a negative keyword would have caught in the first week.

The same data, cut by match type

Every search term also carries the match type of the keyword that caught it.

Two things in that table.

First, exact match is 2.4× cheaper per lead, at a conversion rate more than twice as high, and it gets 9% of the budget. Phrase takes 91%.

Second, and this is the part that closes the argument: the CPCs are within 12% of each other. $11.43 against $10.22. Exact match isn't cheaper because it bids lower. It's cheaper because it buys better queries at roughly the same price per click. The entire difference is query quality.

The broad row is worth a footnote. Twelve broad-match keywords were enabled across these campaigns and recorded essentially no spend in the window. Broad didn't lose the comparison; it never entered it. Which is its own small finding about how little the platform actually reaches for broad match when phrase is available.

Nine years of the same pattern

Ten days could be a fluke. So we pulled the account's entire history, $25.7M of search spend across nine years, and grouped campaigns by the match type in their name.

Exact produced 84% more conversions from 31% more spend, and was 29% cheaper per lead. Across 173,859 conversions. The gap in the ten-day window is sharper than the lifetime gap, 2.4× against 1.4×, which suggests it has been widening as close-variant matching has expanded.

The query-length view

A different cut of the same ten days, by how many words the searcher typed:

The very long tail, eight or more words, converts at 7.2% and costs $169 per lead. It gets 3% of the budget. But notice that the relationship isn't monotonic: 6–7 word queries are the worst bucket. The simple "long tail converts better" claim is wrong here. Only the very long tail wins, and the reason is probably that eight-word queries are specific enough that close-variant matching can't drift far from them.

Why phrase match leaks

Since 2021, phrase match in Google Ads has absorbed what used to be broad match modifier. The official definition is that a phrase keyword matches queries that "include the meaning" of the keyword. In practice:

The last three are real search terms from this account that matched "solar battery" and cost money. Whether they're relevant is a judgement call. Whether the advertiser chose them is not: they didn't.

What to do about it

The lazy fix is "switch everything to exact match." That's wrong in the other direction: you'd lose the 72 conversions the close variants produced, and you'd stop discovering the queries that should become exact keywords. The right fix is a process, not a setting.

Over time, step 3 moves the converting traffic into exact match where it costs $177 instead of $423, and step 4 removes the waste. The phrase keywords become discovery instruments rather than the main spend line.

How to measure it in your own account

You need the search terms report with three columns: search term, keyword, and match type. Then, for each row, a simple test, every word of the keyword (ignoring the match-type punctuation) appears somewhere in the query.

keyword_words = keyword.lower().strip('"[]+').split()
query_words = query.lower().split()
is_literal = all(w in query_words for w in keyword_words)

Sum spend and conversions on each side. If your close-variant share is above 50% and its CPL is materially higher than the literal share, you have the same problem this account had, and the fix above applies.

The principle

Match types are a contract about what you're buying. Every year that contract has been rewritten in the platform's favour (exact match in 2018, phrase match in 2021), and each rewrite has moved more of the decision about which queries you buy from you to the algorithm.

That's not automatically bad. The algorithm found 72 conversions this advertiser wouldn't have written keywords for. But it charged 45% more for each of them, and it charged $1,244 for a keyword that found nothing at all. The only way to know which is happening in your account is to check, and almost nobody does, because the interface doesn't show it. The search terms report does.