Yandex’s built-in antifraud: what it catches
Contents
Share
Yandex Direct’s protection from invalid traffic works at three moments: before the impression, at the click, including through a captcha, and after the visit. This is the account’s built-in antifraud, not a robot label in Metrica. What an invalid click is, and why the money is not always returned in the same hour, is covered in invalid clicks: what they are and why they are not always refunded.
The antifraud page does not promise that such traffic no longer reaches the site. Below is what the filter does with payment, what it is not, and where the help page itself leaves a gap.
Three moments of one filter
Before the impression, the ad may not be shown if the system is confident it is facing a bot or a fraudster. At the click, a suspicious visit may get a captcha. After the visit, statistics go through what already happened once more.
The help page describes this as an update for campaigns on Yandex Advertising Network and on search. Previously the algorithm examined behavior after the click and subtracted invalid impressions, clicks, and conversions from statistics, but it did not stop that traffic from reaching the site. Now it also lowers the risk of the visit itself. Lowering the risk is not the same as closing the entrance. The next paragraph on the page says outright that fraud can still reach the site.
A captcha on a suspicious click is one technique of this filter, not a separate product and not a guarantee. Confidence that “this is a bot” is required to refuse the impression. While that confidence is absent, the ad may be shown, and the review left to the click or to the repeat check.
Online does not charge, offline can freeze the amount
Clicks found invalid in online filtering are not paid for. Statistics are then checked again. Some clicks already counted can also be filtered, the money for them is returned, and a notice arrives at the campaign’s mail.
While the check is running, Direct blocks the amount of the click’s cost, but the interface does not show this as a charge. A block is not an expense line. It is not visible where a charge usually sits, and that is why the balance can behave unlike the operation history.
If the blocked amount reaches the balance on the campaign account, impressions stop. They resume when the invalid clicks are unblocked. A stop at that moment is not the end of the budget and not a fine. It is the frozen amount hitting the balance.
The help page suggests estimating potential spend as the number of filtered clicks multiplied by the average click price. A worked example, not a formula for campaign loss: 40 filtered clicks at an average price of 25 ₽ give 1,000 ₽ to compare with the balance. If the balance is below that estimate, impressions can stop until the check unfreezes the amount. The 40 and the 25 ₽ are arbitrary figures of the example. The help page does not call them a typical share.
Metrica does not stand in for this filter
Robot detection in Metrica and Direct’s antifraud are different systems. A conclusion from one is not substituted for a conclusion from the other. A robot in a Metrica report does not mean Direct found the click invalid and returned the money. A click Direct filtered online may not land in the statistics you are comparing visits against at all.
The news of 23 September 2021 calls Direct’s antifraud a separate tool: an update to robot detection in Metrica does not affect it. Adjusting a Metrica report and waiting for the Direct account to charge differently is not something that news allows. The date of a Metrica update and the date of Direct’s filter live apart.
| Question | Direct antifraud | Metrica robots |
|---|---|---|
| When it looks | before the impression, on the click, after the visit | its own system, not these three moments |
| What it does with money | an online click is not paid; a return is possible later | its conclusion is not substituted for Direct’s filter |
| Are the updates linked | news of 23 September 2021: a Metrica update does not change the filter | a separate tool |
What the page does not promise
The filter lowers the risk of the visit and still allows that fraud will reach the site. A subtraction from statistics is not the same as the absence of a visit. Previously this was said outright: invalid impressions, clicks, and conversions were subtracted, and the traffic still reached the site. A refusal to show the ad when the system is confident, and a captcha on a suspicious click, were added. The paragraph that fraud can still reach the site remained.
While the offline check is running, the frozen amount is not visible as a charge. Looking for it in the spend column and not finding it is what the page describes. Comparing it with the balance makes sense as the estimate “filtered clicks × average price”, and that estimate is not a calculation of campaign loss.
The captcha and the refused impression are not described as a full list of what the filter can do. The page names three moments and two newer techniques. Anything absent from that description does not follow from it.