Skip to content
← All notes

Notes

Exact match isn't exact anymore

The word “exact” is carrying more than it can back up. If you have ever bid an exact-match keyword expecting to show only on the term you typed, your search terms report has probably already disappointed you. Exact stopped being exact around 2019, and the gap between what the label says and what the keyword does is where a surprising amount of budget leaks out.

What exact used to mean

Before match types were loosened, the labels were literal. Exact [oak tables] bought the term, its singular and plural, and minor misspellings, and almost nothing else. Phrase "oak tables" bought queries that contained that phrase in order, like “buy oak tables” or “cheap oak tables”. The label and the behavior matched, so you could read an account by its keyword list and know roughly what you were paying for.

That is the same loosening, viewed from the other side, that makes your negative keywords keep leaking. Regular keywords got wider; negatives did not.

What exact means now

Today exact matches what Google calls close variants. Those are queries it judges to carry the same meaning as your keyword. That covers the singular and plural, reordered words, implied and function words, and outright synonyms. Google’s own documentation uses the example of [lawn mowing service] matching a search for “grass cutting service”. Different words, same intent.

The trouble is that Google decides what counts as the same meaning, and its bar is looser than yours. [oak tables] reliably picks up “oak table”, “tables made of oak”, and “solid oak tables”, then keeps drifting. It can stretch to “oak effect tables” or “oak veneer tables”, shoppers after the cheap laminate look rather than the solid oak you sell. Each hop is defensible to Google on its own, but the sum is a keyword that looks tight in the interface and spends like something much broader, on searches a tier or two off from what you actually offer.

Phrase drifts toward broad

Phrase loosened the same way, and in competitive, high-CPC niches it drifts the furthest. The more expensive the core queries get, and the less of that literal phrase you can actually win, the harder the bidding algorithm is pushed to explore adjacent queries instead.

When your bid can't win the core query, the algorithm spends it on cheaper searches Each bar is what a search costs to win. The core query costs more than your bid, so you can't win it. The adjacent and competitor searches cost less than your bid, so the algorithm spends your budget there instead. WHAT EACH SEARCH COSTS TO WIN, VERSUS YOUR BID your core query adjacent search adjacent search competitor term your bid spend flows here
Each bar is what a search costs to win. Here your bid sits below the core query's price, so you lose it and the budget flows to the cheaper adjacent and competitor searches the algorithm can win instead.

It is not only a question of bid. Even when your bid clears the core query, the algorithm keeps spending part of the budget testing adjacent searches, and the less conversion history the account has, the more it explores instead of exploits. A new or thin account drifts the most.

A phrase keyword like "managed it services" starts surfacing on “it support near me” and, worse, on rivals’ brand names. In the roughest niches phrase behaves like old broad. You bid a tight phrase and your search terms fill up with competitor names you never chose.

How wide each match type matched in 2019 versus today For each match type the upper light bar is how wide it matched in 2019 and the lower dark bar is how wide it matches today. Every type grew. Today exact reaches synonyms and same-meaning rewordings, and in competitive niches phrase stretches close to where broad used to be, which is how competitor names reach your search terms. HOW WIDE EACH MATCH TYPE MATCHES 2019 today narrower the whole category Exact [oak tables] synonyms, rewordings Phrase "oak tables" competitor names Broad oak tables
Light bar: how wide each type matched in 2019. Dark bar: today. Exact now reaches synonyms and same-meaning rewordings, and in a competitive auction phrase stretches close to where broad used to sit, which is how competitor names end up in your search terms.

The keyword is a hypothesis now, not a contract

The practical consequence is that you can no longer read an account by its keyword list. The keyword tells Google roughly what you want. The search terms report tells you what you actually bought. Those became two different documents the day the match types loosened, and the distance between them grows with how competitive the niche is.

An account that looks disciplined at the keyword level, all exact and tidy phrase, can be full of competitor names and off-intent searches once you open the search terms. Nothing in the keyword column tells you that. The report is the only place it shows.

The fix

Read the search terms report as if it were your keyword list, because functionally it now is. That is where the account actually lives, and staying on top of topical drift is most of the job.

What holds the drift in check is conversion data. When an account has enough conversions, Smart Bidding learns which of those loose variants actually pay and pulls its bids back on the ones that don’t, so broad can run fine and you let it. When it does not have that data yet, or the CPCs are high enough that one low-intent click costs you $90, you tighten back toward exact so the drift cannot bleed you while the algorithm is still guessing. The right match type is a function of how much the account knows and how expensive a wrong click is.

The first principle is to configure the campaign and its structure so you already know what it will do, instead of running it and finding out. Match type, bid strategy, budget, and how you split the ad groups all decide how wide it actually runs. Set those deliberately and a broad campaign is completely fine, because you built it broad and you know that is what it is. The mistake is not knowing what your own configuration is likely to do, so you manage a campaign as if it were narrow while it spends like a broad one.

Negatives are the part that still behaves either way. The broad-match negative behavior covered in the negatives note applies narrowly and predictably, so it is the lever that keeps a loosened keyword honest. And watch close variants. If a keyword’s conversions are coming from variants rather than the term itself, that is the loosening at work, and it is a signal to either add the good variant as its own keyword or negative the bad one.