Negative keywords in Yandex Direct: a system, not a list

Most advertisers treat negative keywords as a one-off task. You collect a list of banned words at launch, paste it into the campaign, and forget it. Then you wonder why the budget melts, why inquiries keep arriving but none of them are worth calling back, and why the sales team is annoyed about “empty” calls.

A negative keyword list is not a stop-list you finish once. It is a live filter on traffic quality, and it works only while someone maintains it. In Yandex Direct that is truer than in Google Ads, for a reason that has nothing to do with the interface and everything to do with Russian grammar.

The experiment that proves the list is an asset

Take an account where negative keywords have accumulated over several years and remove them all at once. Logic suggests you will release hidden demand and get more customers.

In practice the opposite happens. The number of inquiries jumps, and none of it is demand. People from cities you do not serve. Random queries. “Just looking.” Job seekers, students, competitors’ employees. The business starts burning hours on conversations that will never become a sale, and the cost per real customer goes up while the cost per lead in the report goes down.

That accumulated list was not ballast. It was an asset that had been quietly protecting you.

Which gives the first conclusion, and it is the one that changes how you work: if you think about negative keywords as a list, you have already lost. A list is finite, so people “finish” it. A system gets maintained.

Why Russian makes this harder than you expect

If your team runs Google Ads in English, the negative list you are used to is mostly nouns and a handful of intent words: free, jobs, cheap, DIY. Import that habit into Yandex and it collapses, for three reasons.

Russian inflects everything. One noun has six cases and two numbers. A verb changes by person, tense and aspect. The word for “buy” alone shows up as купить, куплю, покупка, покупаю, покупать. Yandex matches these forms unless you pin the form with an operator, so a negative keyword added in one form may leave five doors open.

Word order carries less weight. Russian speakers reorder freely, so the same intent arrives in many shapes. A negative phrase fixed by order catches one of them.

Translated lists miss the local junk. The words that waste Russian budgets are not translations of the words that waste English ones. They are the names of Russian marketplaces, local competitor brands, regional slang for “used”, and the particular phrasings people use when they are shopping for a job rather than a service.

None of this is an argument for a bigger list. It is an argument for a different method.

Why a blacklist loses to a whitelist

The usual cleaning routine: open the search terms report, find a junk word, add it as a negative. Query by query, word by word. It is an endless race, because junk is generated faster than you can exclude it. You are always one step behind the algorithm.

There is a stronger approach, and it inverts the logic. Instead of listing everything bad, you describe everything good. And you work with individual words, not phrases.

The mechanics:

1. Export the full search terms history of the account, as far back as it goes.
2. Split every query into separate words. Do not normalize them to a dictionary form: купить and куплю are different units here, and that difference is what makes the filter accurate.
3. For each word, check whether it ever appeared in a query that converted, in any campaign, at any time.
4. Words that have taken part in at least one sale go into the allowed list. That is your whitelist.
5. Now keep in the campaign only those queries built entirely from whitelist words.

Read step five again, because it is the whole idea. Not “remove phrases containing a bad word” but “keep phrases where every word has already proven it leads to money”. The first approach excludes one junk word and lets a thousand new ones through. The second excludes everything unproven, at once.

What you see after this analysis is almost always the same shape. More than half of the unique queries bring a tiny share of clicks and take a noticeably larger share of the spend. Those are the queries that cost a lot and return nothing. You can cut them without fear of losing demand: you lose a drop of clicks and recover a visible part of the budget, which then flows to where conversions are already proven.

A whitelist is also several times more compact than a blacklist. It is easier to hold in your head, easier to check, easier to hand to another person. And it does not go stale as fast, because it describes the shape of your demand instead of fighting its endless distortions.

How not to block your own demand

The method has a dangerous side. Cleaning is addictive, and in the heat of it people throw out what was making money.

In audits of other people’s accounts this shows up constantly: competitor queries sitting in the negative list, adjacent formulations that were converting, the customer’s own pain phrased in their words, and occasionally the name of the advertised product itself. Someone closed off a slice of hot demand with their own hands and never noticed.

The cause is that the decision gets made by eye. “This word looks off-target,” into negatives it goes. But intuition misfires even for experienced specialists, because the same word means different things in different niches. So a doubtful word should be checked, not guessed.

The check takes a minute and leans on several independent sources:

• The Yandex results page for that word. What is actually shown to a person, and what intent sits behind the query.
• The Google results page for the same word. Intent sometimes differs between the two engines, and the difference itself is a hint.
• Image results. Visual context often reveals the real meaning faster than text does.
• Deeper phrases. Longer formulations containing the word, in a keyword tool. Whatever they are made of is where the word pulls.

The rule is simple: run the full check only on borderline words. An obvious typo or a clearly foreign topic is visible at a glance and does not deserve the time. Anything that raises doubt is worth looking at with your own eyes before you close a channel. One mistake in the negatives costs more than a minute of checking.

And separately, about revision. A negative list has to be reviewed, not only extended. A target word sent into negatives by accident at launch will quietly cut your inquiries for months, and no report will show it as an error. What you will see is “demand seems weak this quarter”.

Where the expensive junk hides

Some of the waste is on the surface and gets cleaned as you go. The costly part hides where you are not looking.

Competitor and vendor brands. Yandex autotargeting tends to show your ads on brand queries belonging to competitors and manufacturers, and those names are not always obvious to a foreign marketer. They hide deep inside category queries and cannot be collected by hand. Clean foreign brands and domains systematically, not one at a time. And so the algorithm understands who you are, name your own brand explicitly: in the ad headline and text, and in the title and description of the landing page. That raises the chance the system links your impressions to queries about you rather than about the neighbour in your niche.

Hidden search terms. Some search systems hide part of the formulations that produced clicks, and you pay blind. You cannot switch the hiding off, but you can narrow the corridor: pre-emptively exclude broad junk vocabulary that the system considers close to your keywords. This is work done in advance rather than after the fact, and it noticeably reduces the share of off-target traffic inside the part you are never shown.

Over-broad phrases made entirely of target words. Every word looks like yours, but without specifics the phrase attracts everyone. These are caught separately, in exact match, so you cut the broad variant without touching the narrow commercial ones.

The hygiene without which the list rots

Even a well-built keyword set degrades technically over time. The classic case: someone typed a phrase without switching the keyboard layout, so Latin characters ended up inside Cyrillic words. Or diacritics and stray special characters settled into the list. Across thousands of rows, no one catches that by eye.

What helps here is not patience but an automatic check. A regular expression pulls out of every phrase everything outside the expected character set. You expect Cyrillic, digits and markup symbols; everything else gets highlighted. An empty result means the phrase is clean. Something extracted means there is a foreign character worth checking. Five minutes of work instead of an hour of proofreading.

This kind of hygiene looks like a detail, but trust in the data is built from details. A dirty keyword set produces dirty statistics, and on dirty statistics you make the wrong decisions about bids and budget.

What to do with your advertising this week

A short route you can walk right now:

1. Test the asset hypothesis. Open your accumulated negative keywords and ask honestly whether you know what is in there and why. If the list has been maintained by “someone, at some point”, it needs an owner.
2. Build the whitelist. Export the full query history, split it into words, keep the words that took part in at least one conversion, and run only the queries built entirely from them.
3. Run a revision. Find the target words that landed in negatives by mistake and give yourself back the demand you closed off.
4. Close competitor brands and name your own. Systematically, not one by one, and say your brand in the ad and on the landing page.
5. Do the hygiene. Scan the keyword set for layout mistakes and stray characters, so the data can be trusted.

Negative keywords rarely make it into a nice report or a conference talk. Nobody shows a client a cleaned list as an achievement. But this is where budget leaks most often, quietly, while everyone is looking at bids and creatives.

If your advertising brings inquiries and they are the wrong ones, the problem is almost never the bids. It is who you are letting in.

Frequently asked questions

Sources

• Yandex Direct help, keyword operators and negative keywords: yandex.com/support/direct/en/keywords
• Yandex Direct help, search terms report: yandex.com/support/direct/en/statistics
• Russian version of this article on belokrylovo.ru: Минус-слова в Директе: система, а не список