Fraud refers to malicious or erroneous clicks, conversions, or impressions that are intended to drain the advertiser's budget. Yandex.Direct takes the security of our advertisers very seriously, and we employ multilevel anti-fraud technology. This includes both automatic and manual filtration methods. Fraudulent clicks, conversions, and impressions are filtered out at every level. They are not reflected in Yandex.Direct statistics reports, and funds are not withdrawn from the advertiser's account for them.
How fraud protection works
- Robot protection
- At this stage, the system filters out clicks and impressions made by robots. Robots are programs that imitate the actions of an actual user. They can be used to spend an advertiser's budget or ruin the CTR for ads.
- Online filters
- Then, in online mode, the system analyzes every impression and click by checking them along a list of 250 different characteristics. These characteristics include: time of day, season, geography, certain events leading up to the click, etc. The list of these characteristics changes based on long-term studies of trends in user behavior. At this stage, the system filters out clicks and impressions that were determined to be malicious based on the analysis of these characteristics.
- Offline verification
- The majority of fraudulent traffic is filtered out during the first two stages. Then, at stage three, the system analyzes user behavior over time based on a complex of characteristics. The system analyzes approx. four billion events daily using clustering algorithms. This helps us discover and eliminate complex instances of fraud.
Yandex specialists constantly track fraud and improve advertiser security.
Accounting for fraud in statistics
Expenses incurred due to fraud are refunded automatically to the shared account. The impressions and clicks filtered out by the system are deleted from all statistics reports. To see the number of clicks filtered by the anti-fraud system over a selected period, launch the Report wizard and open General statistics or Search queries reports (Invalid clicks line). These figures may change over time if a portion of clicks are considered invalid during future filtrations.
In part, spikes in the number of clicks or impressions may be caused by factors that are not fraud related. The most frequent reasons for spikes in activity are described in the I noticed a spike in impressions, clicks or expenditures section.
Fraudulent conversions are actions aimed at spending the advertiser's budget. The following conversions are not considered fraudulent:
- Repeat conversions within different sessions
Sessions are defined according to one of four attribution models. For example, a user submitted a request during different sessions. This is not considered fraud.
A session on the "Thank you for your purchase" page is set as a goal, but the user forgot to close the page and the conversion was counted twice. This is not fraud either.
To avoid such situations, use verifiers that prevent users from sending the same data again. Either change the attribution model or the goal for optimization.
- Fields filled in incorrectly
- Contact details. An invalid email or phone number specified by the customer aren't fraudulent conversions.
Inappropriate use of fields. For example, a user left feedback in the contacts form or submitted other information instead of a phone number. Filling out a form in exchange for a reward promised on the site is not considered a fraud either.
To avoid such situations, carefully adjust the data format in forms and try not to mislead users.
If you suspect that some conversions were made deliberately to drain your ad budget, contact us using the form.