Most accounts spend more than half their budget on traffic no one deliberately targeted. A study spanning over 4.5 million clicks and 400,000 conversions found close-variant search terms consume more than half of the average Google and Microsoft Ads budget while performing worse than true matches, with Google showing the worst CPA and ROAS of any match category measured — roughly double the cost per acquisition and about half the return of terms that matched the keyword directly. That is not a rounding error. It is the single largest optimisation lever most accounts are leaving untouched.
Close variants are the queries Google fuzzy-matches to your keyword — plurals, synonyms, reworded phrases, same-intent searches — and they apply to every match type with no opt-out. You cannot stop the matching, so the question is not how to turn it off but how to see it, measure it, and claw the losing half of it back. This post puts a number on the waste and then walks the search-terms-report workflow that isolates the close variants bleeding your budget from the ones worth keeping.
The number, and why it is bigger than it looks
Half the budget at double the CPA is the headline, but the shape of the loss matters more than the average. Close variants are not uniformly bad — some convert fine — which means the losing variants are dragging the blended number down while hiding behind the winners. When a match category runs at half your ROAS on average, the worst rows inside it are running at a small fraction of it, and those are the rows you can cut with no downside. The average understates the opportunity because the fixable waste is concentrated, not spread evenly.
It looks worse once you account for what the spend displaces. Every pound spent on a low-converting close variant is a pound not spent on the queries that do convert, and under Smart Bidding it also teaches the algorithm that this traffic is worth chasing. So the waste is not just the direct spend on bad clicks — it is the mislearned bidding signal that keeps buying more of them. This is the quantified version of the hidden wasted spend problem: the leak has a size, and for most accounts that size is over half the budget.
Why close-variant waste stays invisible
The reason this survives in so many accounts is that close variants do not announce themselves. A close variant appears in the search terms report as an ordinary query under an ordinary keyword; nothing flags that it was a fuzzy match rather than a direct one. Unless you are actively comparing each query against the keyword that triggered it, a synonym or reworded phrase reads as a normal, relevant search. The waste is not in obviously junk queries — it is in plausible-looking ones that quietly underperform.
It also cannot be pre-empted at setup, because there is no close-variant setting to tighten. Advertisers who came up when exact match meant exact still carry the assumption that a clean keyword build protects them; it does not. The matching happens at serve time regardless of how carefully you chose your keywords, which is why close-variant control is an ongoing hygiene task rather than a one-time configuration. The only place it becomes visible is the search terms report, after the money has been spent, which is exactly why a regular report cadence is non-negotiable.
Isolate the losing variants in the report
To find the waste you have to read the search terms report with the triggering keyword next to each query, so segment it by keyword and match type first. Add the "Keyword" and "Match type" columns to the report, and now every row shows both the query and the keyword that matched it. A close variant is any row where the query differs from its keyword — a plural, a synonym, a reordering, or a topically related phrase. The search terms report walkthrough covers pulling and configuring the report if you have not set it up this way before.
Sort by cost and read for the close variants that spend real money and convert poorly or not at all — those are the rows carrying the half-budget waste. The pattern to notice is the same one behind exact match degradation: the query has drifted from the keyword's intent, just not far enough to look obviously wrong. For accounts with many keywords, run an n-gram pass over the query column so the wasteful words rise to the top as high-cost, zero-conversion terms instead of making you read thousands of rows by hand.
Claw the spend back without cutting the winners
The mistake here is to blanket-negate close variants, which throws away the converting ones along with the losers. The disciplined version is a two-way split. For a close variant that spends and does not convert, add a negative targeted at the specific wasteful query or the word that defines it, at the tightest level that covers the leak. For a close variant that converts well, do the opposite: promote it to its own keyword so it gets bid, budgeted, and managed deliberately instead of leaking through a different keyword at a price you never chose.
Structure makes both moves cleaner. Segmenting keywords into match-type-specific ad groups — exact in one, phrase in another — means a close variant matched through phrase is visible as a phrase-level event, and you can fence it with a phrase-scoped negative without touching the exact keyword that shares its theme. This is the account-hygiene backbone the negative keyword list structure post lays out, applied specifically to close variants: the tighter your structure, the more surgical your negatives can be, and the less collateral damage you do to the traffic that pays.
Guardrail against over-correcting
Half your budget being fixable does not mean half your budget should be negated. Close variants include queries you genuinely want, and an aggressive negative sweep can strip out converting traffic and starve Smart Bidding of the signal it needs — the failure mode the over-negating post details. The metric that keeps you honest is conversions: only negate a close variant once it has spent enough to prove it does not convert over a window longer than your conversion lag, not on the first zero-conversion week.
Keep a dated log of every close-variant negative — the query or word, the level, and the reason — because you are touching a large share of the account's traffic and the effects take weeks to read. If volume or conversions dip after a negative sweep, the log is how you find the one fence that went too far and reverse it. Done this way, the close-variant problem stops being a vague half-budget leak and becomes a repeatable weekly pass: read the report, split winners from losers, promote the winners, negate the losers, and log both. That routine is where the doubled CPA comes back down.