Click fraud in the Yandex Advertising Network: four scenarios and how to cut them

In ad networks, fraud is not a rare glitch. It is the background condition.

You clean the placement blacklist, and two days later it is full again with new names of the same kind. Individual clicks cost ten times the usual price and bring nothing. Support tickets about fraudulent traffic mostly go nowhere. It sounds unfair, and it is the right place to start from: protecting the budget in the network is your job, not the platform’s.

The reason is structural. The ad system is at once the place where ads run, the party that earns from the budget being spent, and the moderator of network quality. In a market without strong competition, such a system has limited incentive to fight fraud inside its partner network quickly. So the realistic position for an advertiser is simple: do not wait for someone to solve it for you, learn to recognise fraud by its symptoms and cut it by hand.

For a foreign company this matters more than for a local one, for one unglamorous reason. When you run Yandex from another country, you usually cannot make the calls that check lead quality, and the agency reporting to you shows cost per lead, not answer rate. Fraud hides exactly in that gap.

Below are four typical scenarios and what to do with each.

Scenario one: expensive clicks, no inquiries

The most direct sign of fraud in the network is an abnormal cost per click with nothing to show for it. Certain placements charge amounts an order of magnitude above the typical, and no qualified leads come from them. That is not “an expensive niche”. That is the budget being spent on your behalf.

The defence here is manual and regular. Sort placements by cost per click and by spend, find the abnormally expensive ones with no conversions, and add them to the exclusion list.

The key word is regular. The list fills up again within days, so a one-off cleaning does not help. This is not a “set it and forget it” task, it is hygiene you maintain for as long as the network campaign runs. In practice, once a week is the minimum rhythm; on a large budget, twice.

Scenario two: cheap leads that look too good

The opposite scenario, and a more treacherous one. One campaign suddenly starts producing inquiries several times faster than usual, and cost per lead falls by a factor of three or four. Instinct is delighted: the offer has landed. Instinct is wrong.

Behind a sudden flood of cheap leads on simple forms, quizzes especially, there is often bot traffic. It comes from bait sites generated end to end by a neural network: text, images and code from one template, differing only in topic. Such placements are produced by the dozen every day and get into the partner network within a week, despite the declared quality review. The contact details in the leads look plausible, and you cannot tell a fake from the form alone.

It is recognised by one move: try to call. A sharp divergence between how cheap the lead is and how often anyone answers is the marker of a bot. The numbers turn out to be non-working or non-existent. So an abundant, suspiciously cheap flow of inquiries from network traffic gets checked by contact quality, not celebrated as a number in a report. If suspicions are confirmed, exclude the source placements and tighten quality control.

An honest caveat: not every rise in cheap leads is fraud. Sometimes it is a good offer or a season. The conclusion comes after checking contacts, not from the price drop alone.

For a company outside Russia this step needs a plan. Somebody who speaks Russian has to call a sample of leads within a day or two. A distributor, a partner, a virtual assistant, anyone. Without that check, a bot flood can run for a month and look like your best campaign.

Scenario three: demand that does not exist

Bots distort more than clicks. They distort the demand data you plan on.

Search volume can be inflated: a bot types the same phrase repeatedly in one fixed set and order of words, raising its visible popularity. Highly competitive category and brand queries in almost any niche get the worst of it. The result is that you build your keyword set on demand that is not really there, and for a foreign advertiser researching an unfamiliar market, that is an expensive mistake, because you have no instinct telling you the number looks odd.

Inflated volume is revealed by a combination of markers:

• One frozen form. A whole cluster of phrases has only one exact-match volume, and it equals the base volume. The phrase lives in one rigid form, which real language does not do.
• A jump from zero. Seasonal history shows the query did not exist, then volume leapt from zero to something noticeable in a short period, with no organic growth before it.
• No depth. The phrase has no nested variety: no range of qualifying words and word forms that live demand always grows around.

Each marker alone proves nothing. Together they are a strong signal of artificial origin. The practical conclusion: do not trust absolute volume figures on competitive and brand phrases blindly. Check them against these markers, and if inflation is confirmed, do not use the phrase as a measure of demand.

Scenario four: aggressive formats that burn quietly

A separate leak is the newer mobile ad blocks. A block appears that, a few seconds after the page loads, pins itself over the content. Responsive ads land in it automatically, and you cannot switch off placement in it completely: the decision belongs to the site, not to you. The result is predictable. Mobile click spend rises sharply and leads do not.

The defence is regular monitoring of the splits. Not once, continuously:

• Watch the device split in network campaigns. A sudden jump in traffic and spend without a rise in leads is the typical symptom of landing in such a format.
• When it jumps, look at devices and ad format first. Look for horizontal display variants of responsive ads on mobile.
• Before excluding a placement, check it by device. If desktop through the same placement does deliver leads, the decision depends on priorities: keep it and accept a higher average cost per inquiry, or exclude it and lose those conversions.
• Build the format’s aggressiveness into your bounce expectations. A user closing one block can trigger another, and a high bounce rate here is not always your landing page’s fault.

What to keep in your head permanently

A short defence summary:

1. Treat protection as your task. Network support is slow for structural reasons. Count on yourself.
2. Clean placements regularly. The blacklist refills. A one-off tidy-up does not work.
3. Check cheap leads by phone. Cheap plus unreachable is a bot marker, not luck.
4. Do not trust inflated volume. Check competitive and brand phrases for the markers before building a keyword set on them.
5. Monitor devices and formats. A spend jump without leads is almost always visible in the device split.

Fraud in ad networks is not defeated once and for all. It is kept under control, like weeds. Whoever walks through placements, devices and lead quality once a week loses a little to bots. Whoever looks only at the total spend figure feeds them for months and wonders where the budget went.

If network spend is growing and inquiries are not, or inquiries arrive and nobody answers the phone, the problem is almost always traffic quality. Working out how much of your budget goes to bots, and how to cut it, is exactly what the free review covers.

Frequently asked questions

Sources

• Yandex Direct help, invalid clicks and traffic quality: yandex.com/support/direct/en/general
• Yandex Direct help, the Yandex Advertising Network and placement settings: yandex.com/support/direct/en/campaigns
• Russian version of this article: Боты и фрод в рекламных сетях