Exact match is supposed to be the match type you reach for when you want control — the one that shows your ad only when someone searches the thing you are bidding on. That contract has quietly eroded to the point where it no longer holds. In 2026 an exact-match keyword can trigger on queries that share none of its words and, in the worst cases, none of its intent. PPC.land documented an exact-match keyword, [best hypoallergenic food for dogs], matching 13 different queries, not one of which contained the word "hypoallergenic" — instead matching things like "dog food for skin allergies." The keyword had been reinterpreted from a literal string into a topic.
For a practitioner this is not an academic curiosity. If your tightest match type leaks into adjacent products and intents, the assumption underneath your whole account structure — that exact match is the safe, controlled tier — is wrong, and the spend proves it before the report does. This post covers what exact match actually does now, how to see the leak in your own search terms report, and the negative-keyword routine that fences it back in without gutting the traffic you want.
What exact match actually does in 2026
Exact match has not meant the literal query since 2019, but the definition of "close variant" has widened every year since. It started with plurals and misspellings, moved to same-meaning reorderings, and then to synonyms and paraphrases. What tipped it over the line this year is that the matching now works on topical categories rather than semantic equivalence — Search Engine Land describes how the match types now behave so differently that phrase match is losing ground to broad because the practical gap between them has collapsed. Exact match sits at the same table now, just one seat over.
The mechanism is that Google's query-matching model reads intent, not text. When it decides that "best dog food for allergies and skin issues" expresses the same underlying need as your hypoallergenic keyword, it serves the ad, even though a human reading the two side by side would call them different products. That is the crux: the model's notion of "same intent" is broader than yours, and it is applied to exact match with no filter you can adjust. The keyword you thought was a scalpel is closer to a category tag.
Why the drift costs you money quietly
The damage is silent because the queries usually look plausible. An allergy query under a hypoallergenic keyword is not obviously junk the way "free" or "jobs" would be — it is close enough that a fast scan of the report waves it through. So the spend accumulates on a slightly-wrong audience: people looking for a product adjacent to yours, clicking, and not converting at the rate your true-match traffic does. Because each individual query is cheap and reasonable-looking, no single row triggers alarm, and the leak hides in the aggregate.
It compounds with Smart Bidding. When the bidding algorithm has conversion data on an exact-match keyword, it will chase whatever the match layer feeds it, bidding up on the adjacent queries if a few of them happen to convert. You end up paying more for traffic that was never the point of the keyword. This is the same dynamic behind broad-match cannibalization, except it is happening inside the match type you trusted to be immune to it.
How to find the leak in your search terms report
The only way to see this is to read the search terms report against the keyword that triggered each query, so segment the report by keyword and match type before you start. Open the report, add the "Keyword" and "Match type" columns, and filter to exact-match keywords. Now every row tells you two things: the query a person typed, and the exact-match keyword that let it through. The leaks are the rows where those two do not describe the same thing. If you are new to pulling and reading this report, the search terms report walkthrough covers the mechanics.
Sort by cost, descending, so the expensive drift surfaces first, and read down the list asking one question of each row: does this query contain the core noun or intent of the keyword it matched? "Best hypoallergenic dog food" under the hypoallergenic keyword is a true match; "dog food for itchy skin" is a leak. Practitioners who watch this closely say the honest answer is to monitor the report from day one of a keyword's life, because the widest drift often appears in the first weeks before the keyword builds enough conversion history for the model to tighten. Do not wait for a monthly audit to catch a keyword that has been mis-matching since launch.
Fence it back in with thematic negatives
Because there is no setting to disable close variants, negatives are the entire toolkit. For each leaking exact-match keyword, identify the off-target word or concept the report shows it matching — the "skin," "allergies," or "itchy" that keeps appearing on queries you do not serve — and add a phrase negative on that concept at the ad group or campaign level. Phrase negatives beat a pile of exact negatives here because the leak is thematic: you are blocking a category of drift, not a list of individual strings, and the category is stable even as the exact queries change wording. This is the same discipline the negative keyword match types post applies to plurals and close variants, pointed now at your positive keywords.
Scale it with an n-gram pass when a campaign has many leaking keywords rather than one. Split the query column into individual words, total cost and conversions by word, and the off-target words that expansion keeps dragging in will rise to the top as high-cost, zero-conversion terms — the n-gram analysis workflow does exactly this. Before you add any negative, check it against your active keyword list so you do not block a word you are legitimately bidding on elsewhere, and add it at the narrowest level that covers the leak so a campaign-wide negative does not starve a different ad group that wants the term.
Structure exact match so drift is visible, not buried
The accounts that survive this without constant firefighting are the ones structured so that drift shows up loudly. Keep exact-match keywords in their own tightly-themed ad groups rather than mixed in with phrase and broad, so that when you read the search terms report you can see at a glance which tier a query came through. When exact, phrase, and broad share an ad group, the report becomes a soup and you lose the ability to tell a genuine exact-match leak from a broad-match query doing what broad match does. Separation is what makes the audit fast enough to actually run every week.
It also changes how you read a drop in volume later. Thematic negatives touch a lot of traffic, so keep a written log of what you negated, at what level, and why, with a date. If a campaign's volume falls, the log is the fastest way to find and reverse a fence that went too far — a discipline worth borrowing from the over-negating work, because the same negative that fences out a leak can, one word too broad, throttle the traffic you meant to keep. Exact match is no longer literal, but a segmented structure plus a watched search terms report gives you back most of the control the match type used to provide on its own.
A weekly routine for keeping exact match honest
Turn this into a standing pass rather than a one-time cleanup. On your normal cadence — weekly for heavier spenders, every two weeks for smaller accounts — pull the search terms report segmented by keyword and match type, filter to exact, sort by cost, and read for queries that do not match their keyword's core intent. Weight the newest keywords and newly-scaled campaigns most heavily, since that is where fresh drift appears before the model settles.
For each confirmed leak, add a phrase negative on the off-target concept at the tightest level that covers it, check it against your active keywords first, and log it. Over a few weeks this pulls your exact-match traffic back toward the queries the keyword was meant to capture, and your CPA on those keywords tightens as the adjacent-intent clicks fall away. Exact match will keep drifting — that is the platform's direction of travel — but a watched report and a thematic negative routine keep the tier doing the job you built it to do.