How to find bots in Yandex Metrica
Contents
Share
Finding bots in Metrica means opening the Robots report for the period where a spike already showed, and reading the names in the rows. It is not a division of the robotness percent, and it is not a search for charged clicks.
The share of visits Metrica classed as robots is covered in robotness in Metrica. The unit here is different, and so is the order of screens.
The period first, then the names
The Robots report guide sets the order for the case when the site starts loading slowly or often answers with HTTP 500. First the Site load report: is there a spike. Then the Robots report for the date or period of that spike. It shows the percent of robot requests to the site and their names. The guide writes the path outright: Reports → Monitoring → Robots.
Without the spike’s period, a month-long table mixes the ordinary crawl with the hour that took the site down. The names are there, and the answer to “who made the spike” is not.
The guide also puts a sample of the table on the Metrica demo counter. Open it before your own account: that is how a named row looks different from the behavior row. The demo counter’s numbers do not carry into your spend.
A row counts data sends, not visits
The metric in the table is called data sends. The guide defines it as the number of page views and service events the tag sent about robots’ actions: a hit call, a non-bounce, and others. Webvisor sends are not in this table. Rows are split by filtering rules. Behavior rules sit on their own row and are not spelled out. The other rows carry the name of a robot that introduced itself in the User-agent.
These are not the visits in the summary and not the clicks in Direct. Suppose that on the day of the spike the search-robot row showed 800 sends and the behavior row showed 200. A thousand sends is not a thousand visits and not a thousand rubles. The numbers are round so the unit is visible. They do not stand in for the percent of requests the report itself shows.
- Teaching example, sends
Why the behavior row has no name, and why a person without JavaScript lands among robots, is covered in bot or human. The counter’s rule does not need a second telling: in the report you read a finished row, you do not rebuild it by hand.
What to do with a name
Robots that introduced themselves are excluded from ordinary statistics at once. The guide on how robots are shown says to look at them in the Robots report: that is how you see whether search robots are coming to index pages. The same page points to how Yandex robots introduce themselves. A name that matches that list means the program introduced itself, not that the click stayed in spend. A name in a row is the purpose the robot announced, and only when the name says so. The catalog of purposes the guides actually name, and the refusal to invent the rest, is in kinds of bots.
An empty behavior row with a large number of sends does not earn a name because the number is large. The guide says outright that these rules are not specified. “So it is a click bot” cannot be copied out of that row.
Before you read a spike as an attack, check the counter’s Address and Additional address fields. Metrica can class visits to addresses that are missing from those fields as robots. That is a setting, not a program.
What the report does not contain
A charged click is not labeled here. The line at which a visit becomes spend is set in what ad click fraud is. A high percent of robot requests, next to real leads and empty invalid traffic, does not measure the budget.
Do not turn on “Filter robots by behavior” for this search. It deletes such visits from the source data, and that cannot be undone. When the option is appropriate, and when it erases history, is covered in the robotness article. It is not needed to read the Robots report: the report guide says robot activity stays visible there even with the filter on, and the extra filtering setting is then already unavailable.
This report does not replace Webvisor recordings: those sends are not in the table. A session that shows movement across the page cannot be opened here.
A check of the visit on the site does not replace the report either. It looks at behavior, the browser environment, the network and the ad-click parameters, and it decides to let the visit through, to give the browser a short task, or to keep it out. If the check does not answer, the visit is let through. It does not print a name from the Robots report. Binetro does not subtract data sends from Direct spend.
A short pass, once the site is already slow:
Open Site load and mark the period of the spike.
Open Reports → Monitoring → Robots for that period.
Read the names apart from the behavior row. Do not translate a name into click fraud when it is about indexing.
Check the counter’s address fields if your own domain landed among robots.
Read Direct spend in Direct. Data sends do not carry over.