Bids in Yandex Direct: what actually decides whether your ad is shown

A bid in Yandex Direct is not a lever for impressions.

You raise it and the ad still does not appear where, by your reckoning, it has to. The reason is that on automated strategies the impression is decided not by your bid and not by how well the keyword matches the query, but by the forecast probability of a conversion. Until that settles in, every conversation about bids runs past the actual mechanics.

Here is what really controls impressions, and where the bid is still your tool rather than an illusion of control.

Why a literal keyword does not produce an impression

A common complaint: “there is not a single ad in the results whose headline repeats my query word for word, why?” Behind it sits a wrong picture of how advertising works.

Having a group where the keyword exactly matches the query, with the whole phrase inside the headline, guarantees nothing by itself. The automated strategy has a different job. It does not show you on every query indiscriminately, informational ones included. It shows you on the queries, and to the people, where the forecast probability of conversion is higher.

From which the main point follows. A particular advertiser at a particular moment may lack the bid, the daily budget, the ad rating or the achievement of the strategy’s targets for the system to choose precisely that perfectly matching group. Instead it shows a semantically close keyword from another group, or autotargeting. Advertising is targeted by ads, not by keywords, and the winner is whoever currently has enough bid, rating, budget and result against goals.

The practical conclusion for a specialist is blunt. Diagnosing a “bad” impression from a screenshot of the results page is pointless. No literal occurrence in the headline is not a diagnosis. What you look at is bids, budget, ad ratings and strategy targets. Those, not the text match, decide whose ad a person sees.

This is also the most common misunderstanding in conversations between a foreign client and a Russian specialist. The client sends a screenshot; the specialist explains that the screenshot shows one auction, for one user, at one moment, and proves nothing.

A separate bid for autotargeting

The autotargeting row lives inside the ad group on the same shared bid as all the keywords, unless you set its bid separately. But autotargeting works on far broader logic: it shows your ad on queries that appear in none of your keywords, including random and thematically adjacent ones.

The result is a distortion. A carefully built target keyword set and a wide, less predictable automatic match run at the same cost per click. Autotargeting accumulates noticeable impressions and spend on weakly relevant queries for exactly the same money as your precision keywords.

The fix is quick and specific. In the group’s bid management, select the autotargeting row and give it a separate, usually lowered, bid. That limits the overpayment for broad traffic without switching autotargeting off as a source of reach.

Negative keywords and search terms cleaning do not replace this measure. Fully cleaning irrelevant automatic matching with negatives usually does not work, so the lowered bid acts as a separate, independent lever.

Micro-conversions as fuel when leads are scarce

An automated strategy learns from conversions. When real monthly inquiries are few, the algorithm lacks data and never reaches a stable mode. The way out is to feed learning temporarily with micro-conversions, intermediate events that happen more often than an inquiry. The methodology has several steps.

Step one. Set up tracking of micro-events through a tag manager. Choose only those that make sense for this site: scroll depth past a threshold, time on site past a threshold, number of pages or cards viewed, a click in the phone field, a click on a quiz or calculator button, revealing a hidden number, adding to cart or favourites.

Step two. Accumulate statistics over a sufficient period.

Step three, which cannot be skipped. Find which micro-event types correlate most strongly with real conversions. If an event predicts an inquiry poorly, it is useless as a signal and the method will not work on it.

Step four, calculating the target price of a micro-conversion. Take the value of one real conversion, usually the net profit from a lead. Work out the transition rate from micro-event to inquiry: how many micro-events ended in a lead. Multiply, and you have the value of one micro-conversion. Then decide what share of that value you are willing to spend acquiring it, and that is your target price.

Step five. Check the weekly volume of the chosen event. It has to happen often enough for the strategy to gather a minimum of target actions per week. Otherwise you are back at the same data shortage the whole exercise was meant to solve. In the settings, specify payment for the most frequent and effective event at the calculated price.

Separately, about overpayment. Paying for several different micro-event types at once charges money for each of them, and one visitor performs them in a bundle during a session. Divide the target price of each event by the number of types, or you pay several times your calculation for one lead.

Is it worth splitting a campaign by city

Another situation where the bid becomes a real lever is different cities inside one campaign.

The auction costs different amounts in each, while the campaign’s effective bid is in practice averaged across the traffic of all cities. If competition differs sharply, that averaging is bad for everyone at once: in cheap cities you overpay per click, in expensive ones you under-collect impressions.

You can assess the benefit of splitting before breaking anything. Collect statistics by city over a representative period: traffic volume, cost per click, share of auctions won. Calculate the campaign’s actual weighted average bid. Compare it with what happens in individual cities.

If the spread between cities is large and one systematically requires a noticeably higher bid to enter, averaging is distorting the picture, and splitting by city with individual bids will give better conditions in each. If the spread is small, there is no benefit, only extra load from managing several campaigns instead of one.

One note: if the regional bid can be fixed with an adjustment inside a single campaign, there is no need to fragment the structure for it. The same effect without extra objects.

What to do with bids this week

1. Stop diagnosing impressions from a screenshot. No literal occurrence in the headline, look at bid, budget, rating and strategy targets.
2. Give autotargeting a separate lowered bid, so broad automatic matching does not buy at the price of your precision keywords.
3. Too few conversions for learning? Set up micro-conversions, check their link to inquiries and calculate the target price rather than guessing it.
4. Paying for several micro-event types at once? Divide the price of each by the number of types.
5. Before splitting a campaign by city, compare the weighted average bid with the situation in individual cities.

A bid rarely controls what you think it controls. It is not a command saying “show me here”. It is one argument the system weighs together with rating, budget and the conversion forecast. Managing advertising on automated strategies means managing those arguments, not moving a bid slider and waiting for the results page to obey.

If bids in your account live their own life and impressions go somewhere other than where you aimed, the free review is where we find what is really driving them.

Frequently asked questions

Sources

• Yandex Direct help, bidding and the auction: yandex.com/support/direct/en/efficiency
• Yandex Metrica help, goals and events: yandex.com/support/metrica/en/goals
• Russian version of this article: Ставки в Директе: аукцион и прогноз конверсии