Exact match is not exact: what the label hides in Google Ads and Yandex Direct

Exact match stopped being exact some time ago. The system serves synonyms, close variants, sometimes a query with one of your words missing altogether, and this is standard behaviour, not a fault. So the label “exact” deserves a different attitude: measure what was actually shown instead of trusting the name.

For a company advertising in Russia the question gets a second layer: Google Ads and Yandex Direct describe matching differently, and Russian itself behaves differently from English inside a search box. Below is where match types mislead, what Yandex offers instead, and what to manage when the label cannot be trusted.

What “exact” actually means now

The mechanics are worth separating. The rules of exact match tolerate a different word order and different word forms, but cut off any new words before, after or between the words of the keyword. Phrase behaviour is wider: it includes everything exact does and additionally lets new words appear around the fixed ones. In Yandex you see this through the fixing operators; in Google through close variants. The outcome is the same: a literal coincidence of query and keyword is long gone.

Two cases are most often mistaken for a malfunction. The first is substitution of equivalent words: the system serves one synonym in place of another, and intent survives. The second is sharper: a word from your keyword can be absent from the query entirely, because by the same logic of equivalence another phrase in the query played its role.

Neither is harmful by itself. It turns harmful in one situation: when a visible share of money goes to such queries.

Russian makes loose matching louder

Here is the part a team coming from an English-speaking market rarely expects. Russian is an inflected language: a noun changes its ending across six cases, adjectives agree with it, and word order is far freer than in English. The same commercial intent can be typed in dozens of surface forms. Any matching system working with Russian has to normalise heavily just to function, and that normalisation is what makes “exact” feel approximate.

Yandex Direct grew up in this environment, which is why it has no match types in Google’s sense. It has operators inside the phrase. Quotation marks fix the number of words. An exclamation mark before a word fixes its form. A plus sign forces a stop word to be present. Square brackets fix word order. Each one addresses a specific way a Russian query drifts away from the keyword, and you apply them word by word rather than choosing a label for the whole keyword.

This gives more control than Google’s current exact match, and it is easy to overuse. A keyword set imported from Google Ads with every phrase wrapped in quotation marks produces a campaign that barely spends, because real Russian queries almost never coincide word for word with a keyword. Direct also has its own version of the Google drift: synonyms and close forms are matched by default, and autotargeting in search campaigns can no longer be switched off completely, only narrowed by query category. Operators and labels are wishes. The search terms report is what happened.

Measure the share, not the incident

This is the main shift in thinking. The question is not “do substitutions and dropped words occur in my account”. They occur in everyone’s. The question is “what share of impressions and clicks falls on them”. In practice it is often close to zero. Until you count it, you do not know.

It is counted with a pivot table at the level of words. From the search terms and from the keyword you extract individual words. To collapse hidden duplicates into a single row, you normalise: remove punctuation, reduce each word to its dictionary form, delete repeats inside a row, sort the words alphabetically. A separate view shows every place where new words appeared in the query that the keyword did not have. After that you see a specific figure of spend on substitutions and omissions rather than an abstract “matching worked differently”. With the figure in hand you decide: tolerate, add negatives, or rewrite the structure.

For Russian the normalisation step is where most attempts die. Lowercasing and stripping punctuation, which is close to enough for English, leaves Russian words in a dozen forms that a spreadsheet treats as different strings. You need a lemmatiser. Without it the pivot shows noise, the share looks larger than it is, and the reaction is usually a wave of negatives that also cuts good traffic. The raw material exists on both platforms as an exportable search terms report. The work is in the normalisation.

Informational queries: broad match will not rescue them either

The other side of blurred matching is the illusion that a broad type can scoop up any traffic. There is a whole layer of queries where paid search barely enters, however many keywords you point at it.

These are informational and multimedia queries: the person wants an answer, a file, a photo or a video, not a purchase. One component of the quality score in Google Ads is expected click-through rate, calculated across the whole industry rather than only your account, and it carries real weight. On informational queries people click ads rarely, whoever placed them. Google sees this low industry-wide click rate and deliberately holds back paid impressions: an irrelevant ad damages the results page, and one-off revenue from a click is cheaper to the platform than the user’s loyalty.

The conclusion is harsh. Dragging informational and multimedia phrases into search campaigns is pointless: they produce close to zero traffic and eat the time spent collecting and maintaining them. Match type has nothing to do with it: there will be no impressions on exact or on broad.

The Yandex side behaves the same way in practice, for the same commercial reason. I treat this as an observation from accounts rather than a published rule, and it holds firmly enough to act on. For a foreign company it matters because Russian keyword tools return large volumes of informational phrases, and a keyword set translated from English tends to keep them. A sort by intent before launch saves a month of confused reporting.

Where matching is powerless, audiences help

Since a match type can neither catch all the traffic nor cut off all the excess, part of the control is moved outside, into audiences and lists. A good source of both is the domains of competitors.

The logic runs like this. From the results pages of Google and Yandex and from the ads on them, you collect the domains that appear for industry queries and sort by frequency: the probable leaders float to the top. Then the list is filtered down to direct competitors: from each domain you take the meta tags, reduce the words to dictionary form and check whether the main words of the niche are present. No target words means the domain leaves the list.

Filtered domains work in two ways. As a source of semantics: brand keywords for these domains go into search campaigns and into networks. And as a ready-made targeting criterion: in Google Ads these are the sites entered into custom audiences by intent. As an allowed list of placements they give minimal reach, so that is a last-resort method. You end up managing not the match label on a keyword but who you appear to and next to whom.

Yandex Direct has its own route to the same idea: under interests and habits, an audience can be built from the sites people visit, and competitor domains go straight into it. The frequency count also tells a foreign company something a keyword report never will: who the real market leaders are in Russian search, which is rarely the set of names from the English-language industry press.

Structure follows demand, not guesses

One last thing. If matching cannot be taken at its word, structure of keywords and site cannot be built on guesses either. Word-by-word analysis of the collected semantics helps here.

The technique: across the whole set of phrases and real queries you count the frequency of individual words, tag each word with a meaning label using regular expressions (attribute, filter type, category) and rank the labels by the total frequency of the words attached to them. The frequency of a word in real demand is a direct signal of how much that attribute matters to people. This ranking decides which filters, tags and menu items to show first and which sections to create at all. One reservation: frequency speaks about demand, not always about commercial value, and those two dimensions are checked separately. A structure derived from data holds up better than any assembled by eye.

For a Russian site this has an extra payoff. A translated site inherits English navigation, built around how buyers in another market phrase their need. Word frequencies in Russian demand often rank attributes in a different order. Rebuilding the filters around the Russian ranking usually does more for paid traffic than any amount of operator tuning.

What to do about matching, especially if you sell into Russia from abroad

• Stop reading “exact” literally. Check your own search terms for where the system substitutes synonyms and drops words, on both platforms.

• Do not port Google match types into Direct as if they were the same thing. Start with plain phrases, read the first week’s search terms, and add operators only where a specific drift costs money.

• Build a word-level pivot with proper lemmatisation and count the share of spend on substitutions and omissions. React to the share, not to the fact.

• Throw informational and multimedia phrases out of search semantics before launch: there will be almost no impressions on them.

• Collect the domains of direct competitors and put them into custom intent audiences in Google and into site-based audiences in Yandex. Early on, this list stands in for the brand demand a foreign company does not have yet.

• Derive the priority of filters, tags and site sections from word frequency in real Russian demand, not from the English site map.

If you want the search terms report of an existing account read this way, start with a free review: you send the export, I show where the money goes. How this fits into a full paid search setup is described on the Yandex Ads page.

A match type is not a lock on the door. It is a request to the doorman. He will understand you approximately and let similar people in. Your job is not to argue with the label but to look at who came through in the end, and to count what the extra guests cost you.

Frequently asked questions

Sources

• Russian version of this article on belokrylovo.ru: Соответствие в Google Ads: почему точное не точное