One ad for the whole keyword set is a leak in Yandex Direct

Pull the ads of your competitors in any category and read them in a row. You will see the same picture almost everywhere: thousands of advertisers running one headline and one text across their entire keyword set. Someone searching for a cheap corner sofa sees the same ad as someone searching for a premium leather one. One message for everybody, because one message is easier to write. This is good news, because the gap it leaves is where differentiation lives, and almost nobody takes it.

The reason for the uniformity is mundane. Writing a separate text for every cluster of meaning takes longer than writing one universal text. At launch the advertiser saves the effort and pays for it later: the ad sounds like “this is about me” to no one. And relevance to a specific intent is precisely what makes a person click on you rather than on the neighbour.

For a company advertising in Russia from abroad the leak usually runs twice as deep. The Russian campaign is built from an English one, so the single universal ad arrives already translated, with claims chosen for another market, and is then spread over a Russian keyword set it was never written for. The fix is the same as for a local advertiser. Only the order of work differs.

Relevance is meaning, not matching letters

The first instinct when adding relevance is to switch on keyword insertion so the headline repeats the user’s query word for word. In Google Ads that is the keyword insertion tag. In Yandex Direct the feature is called a template: a pair of hash marks around a default text, and the keyword takes its place when it fits. The logic seems sound. The closer the headline matches the query, the higher the click-through rate.

In practice it is nowhere near that simple. Compare, inside one ad group, a dynamic headline with insertion against a static headline built on strong arguments, and the static one wins often enough to question the habit. Click-through rate is driven by content as much as by matching words: by the arguments standing next to the keyword in the headline. A person clicks because they read a reason to choose you, and seeing their own word is a weak reason.

Most queries are shorter than the headline limit, so after the keyword is inserted there is nearly always room left for more words, and those added meanings move the click more than the formal match does. And there is a further catch. Insertion puts in the keyword phrase you wrote, not the query the person actually typed. Real queries differ from keywords in word count and word order, so the “hit” on a live query may never happen.

Why insertion is weaker in Russian than in English

Two local details make this worse in Yandex than in Google.

The first is grammar. Russian keywords are usually written as strings of base forms, the way they come out of keyword tools and go into the account: диван угловой недорого купить. Insert that into a headline and it reads like a search string dropped into a sentence, not like something a company would say. An English keyword such as “buy corner sofa” survives insertion. The Russian one does not, because case endings and word order carry meaning that the keyword form throws away. Russian readers see the seam immediately. A headline written by hand for the cluster reads as a sentence, and that alone changes how the ad is perceived.

The second is autotargeting. Yandex keeps autotargeting switched on in search campaigns, so an ad group shows for queries that were never in its keyword list. For those impressions there is no keyword to insert, the template falls back to the default text, and the headline drifts still further from what the person typed. The more of your traffic arrives through autotargeting, the less insertion does for you and the more the static arguments carry.

None of this means insertion is useless. It means insertion cannot substitute for a thought-out structure. The priority is grouping phrases by meaning into clusters, then writing for each cluster a text that answers that intent. “Buy cheap” and “buy premium” are different conversations. Running both through one ad means losing both.

Differentiation by cluster does not take heroic effort

The obvious objection: writing dozens of different ads by hand is work nobody has time for. True, if every text is composed from a blank page.

If the keyword set is already sorted into clusters of meaning, and you hold a bank of selling claims, writing ads turns into distributing ready arguments across groups rather than composing from scratch. The price cluster gets claims about price and value. The quality cluster gets claims about materials and warranty. The urgency cluster gets claims about stock and fast delivery. You are not writing a hundred texts. You are laying out one agreed set of meanings by address.

The current ad format in Yandex Direct helps here, provided you point it in the right direction. A combinatorial ad in a unified performance campaign takes several headlines and several texts and assembles the combinations itself. The temptation is to load one big set of variants and let the system sort it out. That is the uniform ad again, in more expensive clothing. The combinatorial slots are for variants of the same cluster’s argument, so that the system chooses between two ways of saying “in stock, delivered tomorrow”, never between “cheap” and “premium”.

For a foreign company this changes the order of work. The claim bank comes first, and it is built in Russian with a Russian reader, from the offer as it stands in Russia: which payment options exist here, which delivery terms, which warranty. Only then is it distributed across clusters. Translating the English ad and then splitting it is the wrong way round, because the claims themselves may not survive the border. Free shipping means something different once the parcel has to clear customs.

This is the point of differentiation. While the competitor shows everyone the same thing, you show each segment its own argument. On the same budget your ads sound more relevant, and so they collect the click and the inquiry where the neighbour collects a scroll.

Do not hand the ad copy to a language model

The next temptation follows naturally: feed a language model the competitor ads and ask for finished texts. The temptation is understandable and dangerous.

The model does not know the priorities of your business. Out of an array of other people’s claims, a handful apply to your company. The rest are either irrelevant or contradict your offer outright. The model will not tell a primary advantage from a secondary one, and it will not understand that instalments apply to one product group, do not apply to another, and are replaced by a different payment option for a third. Worse, the model is good at adding what you do not have: a condition that exists nowhere on your site. Such an ad at best fails moderation, at worst misleads a person and damages trust.

In Russia the moderation part has teeth. Yandex asks for documentary proof behind superlatives and comparisons, because Russian advertising law treats an unsupported “best” or “number one” as misleading. A model trained mostly on the English-language web produces those words by reflex. A foreign company that lets it write Russian ads will spend its first weeks arguing with moderation over claims nobody in the company ever made.

The working role of the model is more modest and more useful. Not “write me the ads” but “rephrase this already agreed and verified claim”, or “give me draft variants from this reliable set of facts”. People first reduce the array to a short verified list of meanings, and only then the model helps with wording. The final selection and the check against the real offer are steps no automation removes.

Where the easy click-through gain is

While competitors compete on text, part of the click-through rate sits in formats, and that is the cheapest lever by effort. An image attached to a text ad makes it stand out on a results page of solid lines of text. In my experience the gain in click-through rate from it can be a multiple.

It takes little effort. Yandex Direct picks some formats automatically from the site, and the interface offers a library of ready images that already meet the platform’s technical requirements. You can stand out visually without a photo shoot. The ad with a picture on a text results page almost always draws more attention than the same ad without one.

Visibility and relevance work together. The image stops the eye, and the text written for the cluster answers the intent. Either alone is weaker: a good picture over an unaddressed text collects a click and no inquiry, and a precise text without visibility risks going unread.

What to do with your ads this week

A short route:

1. Look at yourself through the customer’s eyes. Take several queries that differ in meaning and check whether they show one ad. If they do, you blend into the market.
2. Tie clusters to claims. Sort the selling arguments into groups of meaning and write text for each intent instead of one for all. If you sell into Russia from abroad, build the claim list in Russian from the Russian offer before you split it.
3. Do not pray to keyword insertion. A strong static headline with an argument often beats a dynamic one, and in Russian the grammar works against insertion. Test it on your own data.
4. Keep the language model on a short leash. Let it rephrase verified claims rather than invent advantages. Check facts against the site and superlatives against what moderation will ask you to prove.
5. Switch on images. It is a cheap and fast way to raise click-through rate on the text results page.

An ad is the meeting point of your meaning and someone else’s intent. When one text hangs over every intent, the meeting does not happen: the person does not recognise their query in the ad and moves on. Differentiation often needs no new offer. It is enough to stop talking to everyone the same way.

If the campaign collects impressions and few clicks, or clicks and no inquiries, the cause is often a text that misses the intent. Where exactly your ads lose the person is something I can show on a free review. How to bind the keyword set, the texts and the structure into a campaign that brings inquiries is on the Yandex Ads page.

Frequently asked questions

Sources

• Russian version of this article on belokrylovo.ru: Объявления: один текст на всю семантику это слив