Clicks are coming in, the budget is being spent, but there are no leads. The usual reaction is to redesign the website. That costs another month, because the problem was somewhere else. Below are ten reasons for low conversion in Yandex Direct, the signs that reveal each one in reports, and the order in which it makes sense to check them.
First, make sure people are clicking your ads
Fraudulent traffic consists of bots that click ads and imitate visitor behavior. They have one goal: to burn through the advertiser’s budget so the placement owner gets their share. These “visitors” have no intention of buying anything.
Worse than the money lost is the wrong conclusion. An advertiser sees dozens, sometimes hundreds, of visits a day in the analytics, but no leads, and decides the problem is the website. Then they start changing something that’s working just fine.
You can check this in a couple of minutes: open this in any campaign: Statistics → By placements and set the date range. The first sign is the addresses themselves. Fraudulent placements have long strings that don’t look like website domains; they usually belong to mobile apps. An address ending in yandex.ru is no guarantee either—placements like these show up in the list, too.
The second sign is more reliable: the click-through rate for a placement. In Direct campaign management, a click-through rate of around 3% is generally considered the upper limit for normal traffic. A CTR of 11%, 23%, or 25% after several dozen clicks means the placement is almost certainly fraudulent. In one account we reviewed, more than 5 000 rubles were burned on placements like these, with no conversions and a high average cost per click.
Two or three conversions from a placement with a 20% click-through rate aren’t a reason to keep it. There may have been real people among the visitors, but you can’t tell them apart from the rest, and you’ll have to pay for every click.
Disable placements in the same report: check the box next to the one you want, choose “Block impressions,” and apply. No one will refund the money already spent, so check the report from the first few days, not after a month of running the campaign.
What to do at the start
There’s no universal filter against fraud. One effective approach is fairly drastic. Fraudulent placements are almost entirely in the Yandex Advertising Network and almost entirely on mobile devices, so while the campaign is learning, set adjustments of −100% for smartphones and separately for tablets. This completely excludes impressions on those devices: Direct adjustments range from −100% to +1200%, and minus one hundred means they’re disabled altogether.
The obvious objection is that 70–80% of traffic comes from mobile devices. But adjustments aren’t meant to stay in place forever. The campaign learns on desktop, reaches stable conversions, and then the adjustments are gradually removed while you monitor how much junk traffic comes in. A trained campaign catches much less of it than a new one: for a newcomer with a small budget, the risk of picking up fraudulent traffic in the network is estimated at over 90%.
Ready-made lists of fraudulent placements from public sources can help, but they quickly become penny-pinching. There are huge numbers of these placements, and new ones appear faster than they can be added to the lists.
How much data do you need to draw conclusions?
Some decisions in Direct are made before there’s enough data to support them. Twenty or thirty clicks tell you nothing, but people draw conclusions from them in both directions.
False positive: two leads from twenty clicks look like a 10% conversion rate, so the advertiser increases the budget expecting the same ratio. Over time, the figure turns out to be different, and the money has already been reallocated.
False negative: no leads after twenty or thirty visits, so the campaign gets switched off. For many industries, that’s a perfectly ordinary stretch of results; conversion rates fluctuate widely.
A good rule of thumb for drawing conclusions: around 300 clicks, or a week to a week and a half of continuous campaign activity. No leads after 100–300 clicks already points to a problem, with one caveat: rule out fraud first, or you’ll be led down the wrong diagnostic path.
A disconnect between the search query, ad, and landing page
A one-word keyword looks like a way to capture all the search demand, but it captures the entire query tree. An advertiser working with OSAGO sets “OSAGO” as a keyword and expects searches like “buy OSAGO” and “apply for OSAGO.” But it also brings in “check OSAGO,” “OSAGO policy number,” “reviews,” and searches containing competitors’ service names. None of these people will submit a lead. The relationship is consistent: the higher the search volume, the larger the share of its query tree that is irrelevant.
The other half of the same problem is what happens after the click. The search query, ad, and landing page should follow on naturally from one another. For the query “buy a Stels bicycle,” that looks like this:
- the connection works: the brand is mentioned in the ad headline and text, the click takes you to a catalog of bicycles from that brand, and all that’s left is to choose a model;
- the connection is broken: the ad is generic—“bicycle shop”—and the click takes you to a page where you have to find the section with the right brand yourself; sometimes those bicycles aren’t even in stock.
The difference isn’t cosmetic; it shows up in the number of leads. The more finely you break down keywords and the more precisely you match ads and pages to them, the higher your conversion rate compared with competitors who send all their traffic to one generic page. This kind of segmentation is possible in almost any industry—in the experience of specialists who have managed campaigns for years, that’s roughly 95% of niches.
A weak landing page: what the data shows and what only seems to be a problem
Conversations about landing-page conversion usually drift into button colors and image choices. These things do have an effect, but it’s worth looking at them after you’ve found the major problems—and those show up in Metrica reports.
Start your diagnosis here: Reports → All reports → Direct, summary. Look at three metrics:
- Bounce rate. Above 30% is already a reason to investigate. There are usually two possible causes: fraud or a page that doesn’t address the search query.
- Pages per visit. For a multipage website, the higher it is, the better.
- Time on site. Fifteen seconds combined with a high bounce rate and low pages per visit gives you an immediate answer. A lead from a site like that would be a fluke, not a consistent result.
Next, go to the “Behavior” section. Session replay is its best-known tool, but also its most inconvenient: watching recordings one visitor at a time takes a long time, and the conclusions are limited to individual cases. Heatmaps are faster.
Scroll map
It’s easy to set up: specify the page in question (people coming from ads don’t necessarily land on the homepage), turn on robot filtering by behavior, and set the date range. The least-viewed areas are highlighted in blue and the most-viewed in red.
The value of the map is that it challenges the site owner’s assumptions about their own website. A typical finding: visitors skim past the testimonials block—which was added to reassure hesitant buyers—without stopping. That means it isn’t working as a selling point, yet it takes up space near the top of the page, so the block should be removed or moved lower down.
Click map
It shows whether people interact with elements that seem prominent. A button that “no one could miss” often gets just a handful of clicks. No clicks on a key element means a direct loss of conversions, and fixing that is faster than redesigning the entire landing page.
A bid works like a range, not a lever
Bids are often treated as a control: pay more, get more impressions and leads. In practice, going too far in either direction hurts.
Bid too low
Systematically shaving a little off the bid for every keyword in every campaign gradually reduces impressions and visits, sometimes stopping them altogether. Then junk traffic starts getting mixed in. The idea is this: people searching for the same query have different needs; Yandex has data on who’s closer to buying, and impressions for those users go to advertisers willing to pay. Those trying to save money get what’s left.
You won’t find this explanation in the official documentation. It’s based on years of observation across the writer’s own and clients’ accounts: the mechanics can’t be verified externally, but the effect is repeatable across campaigns.
Bid too high
The opposite imbalance doesn’t bring in leads either. A high bid doesn’t get you into more of the right auctions; it means the budget needs to be spent. The system starts experimenting and brings in audiences outside the campaign’s targeting—in effect, anyone who clicks the ad. More clicks, no leads.
Every niche and traffic type has its own effective cost-per-conversion range, and going above the upper limit makes results worse. The official materials don’t put it this way: the documentation only says that the set value serves as a reference point for the algorithm and that the actual cost may differ from it. So the explanation above is a practitioner’s assessment, not a description of how the algorithm works.
A demand ceiling in the niche
Sometimes the campaign setup isn’t the problem; there are few leads because there’s little demand. A case in point: registration of pesticides and agrochemicals. In this niche, there are exactly two high-intent search queries: “pesticide registration” and “agrochemical registration.” Wordstat shows around a thousand impressions a month for the first one at its peak, which falls in March. That’s about three searches a day.
An advertiser in this niche got in touch in March, asking why the number of leads hadn’t increased during the busy season. The seasonal pattern is visible on the chart: March and April are higher than the other months, but the increase is from two impressions a day to three. There’s nothing to scale here.
The temptation in this situation is understandable: expand the keyword list and reach people searching for the same thing in different words. Usually that makes things worse. It’s a service with a long sales cycle; people don’t make a decision in a month or even a year, and the number of people who genuinely need it is physically limited. It makes more sense to check search volume before launch and build the campaign around traffic quality, not volume.
Niches where Direct loses not to a competitor, but to another channel
Services offered by individual professionals
Several attempts to run ads for psychologists, coaches, and real estate agents using ostensibly high-intent queries like “psychologist services” resulted in virtually no leads. The problem isn’t the ads. People find these professionals through personal recommendations or on service marketplaces like Avito and Profi, where they can compare several providers side by side using reviews, ratings, photos, and filters.
A solo professional’s website, by definition, shows only their own strengths; there’s no one to compare them with, and that inspires less trust. Direct, Avito, Profi, and social media offer different ways to buy, not different storefronts for the same thing. Real estate agents, coaches, and psychologists should be told about the risks of a campaign like this before it launches, not afterward.
Products that are available on marketplaces
This isn’t about competitors having lower prices or a confusing landing page. The situation is different: the price is competitive, the product is good, but people still don’t buy from you. If the product is easy to order on a marketplace, that’s where buyers will go. The terms are clear, the delivery date is shown, their details are already entered, and they’re used to the payment process. A slightly lower price than the marketplace won’t change that.
An unfamiliar website offers buyers a different scenario: fill out a form, wait for a call, find out the product is out of stock or the price has changed, pay extra for delivery, transfer money to a card with no receipt. After the marketplace boom that began around 2020 and reached almost every product category, this scenario rules out most of the audience: practitioners estimate roughly 80% of the market. Neither campaign optimization nor a high-converting landing page can make up for that. Advertising can work here as an additional channel, but probably not as the main one.
A diagnostic workflow for when there are no leads
Following a sequence saves time: start with the checks that are cheap to do and that most often reveal the cause.
- Placements report: filter out fraud and assess what share of the traffic was real.
- Data volume: have we accumulated 300 clicks or run the campaign for a week to a week and a half? If not, any conclusions are premature.
- “Direct Summary” report: bounce rate, pages per visit, time on site.
- Keywords and landing pages: review the search queries that generated clicks and check where people land after clicking.
- Heatmaps for landing pages: what people scroll past and where they don’t click.
- Bids: check for any skew in either direction.
- Wordstat: is there enough demand to support the volume assumed in the plan?
- Channel: do people in this niche buy through search at all?
The first five points concern setup and can be adjusted within the account. The last three answer a different question: is this channel right for the task? Campaign changes won’t help there; you’ll need to change either the offer or the platform.