Bots in Yandex Metrica: from robot reports to Webvisor
Contents
Share
Bots in Metrica are not a single “robotness” figure. Robots that identified themselves are taken out of ordinary statistics and into the Robots report. Robots Metrica detected from behaviour stay in reports by default. Webvisor can record visits that Metrica or Direct already treats as robotic. The Direct summary counts clicks that passed fraud protection, and its bounces are read as one view shorter than the threshold.
Signs of click fraud are not replaced by this walkthrough. The point is the order of four layers, so a normal robot, a bounce and someone else’s campaign are not treated as the same loss.
Layer one: robots that named themselves
Robots that identified themselves are excluded from ordinary statistics at once. You read them in the Robots report, including when you want to see indexing. If search-robot visits “disappeared” from the main report, a random segment did not cut them: they live in a separate report.
This layer does not answer whether the ad is being clicked fraudulently. It answers whether the site is being indexed, and whether you are looking for a crawler in a report it has already left.
Layer two: behaviour-based robots stay in the numbers
By default, Metrica statistics include visits of robots detected from behaviour. You can remove them from a particular report with a segment. For a comparison, the help text points to the mode “Data: without robots” and the action “Compare with data that includes robots”.
Until you turn the comparison on, bounces, depth and conversion in an ordinary report include this layer. How to find bots in Yandex Metrica is about the search itself. What matters here is that “with robots” and “without robots” are two modes of one report, not two different sites.
Use the same period and the same grouping. Otherwise a gap in bounces mixes robots with the fact that you changed the report.
Layer three: what Webvisor actually shows
Webvisor can record visits that Metrica or Direct already treats as robotic. A recording does not by itself mean the visit was charged or that it stayed in ordinary statistics.
The recording includes what the visitor saw, dynamic content included. A third-party block and an ad are recorded when they are not shown in an iframe. A block inside an iframe does not appear. An empty patch where a widget or banner sat is a limit of the recording, not proof that the visitor “saw nothing”.
The recording has no separate “show robots only” filter. You reach a robotic visit from a report that already marks it, not from a button inside the player. How to look for bots in Webvisor is about watching the recording. This article only fixes that the player is not the same thing as a “without robots” report.
Layer four: the Direct summary and a bounce
The Direct summary opens at Reports → Sources → Direct summary. Its statistics are based on clicks that passed fraud protection in Direct. That is not the whole counter and not the Robots report.
The help text ties bounces in this report to whether the page was left within 15 seconds after a single view. If the counter is not linked to the campaign, visits land in the source “Other campaigns”. Direct itself adds yclid to ad links.
Clicks without leads often starts here: clicks in the summary have already passed antifraud, and the missing leads have another cause — the goal, the form, the offer — rather than “every click is a bot”. Suspicious YAN placements belong in their own report. The Direct summary does not replace them.
A bounce is three conditions at once
A Metrica visit is a bounce only when three conditions hold together: no more than one page view, a duration shorter than the bounce-calculation time (15 seconds by default), and no service “non-bounce” event. The browser sets that event after this time. Because of clock skew and send delay, a visit shorter than 15 seconds may not be a bounce.
Fifteen seconds is the default, and you can replace it. It is not a fraud threshold. A high bounce rate in the Direct summary means that many visits behind counted clicks had one short view and no “non-bounce”. It does not mean those clicks failed protection: the summary is built from clicks that passed it.
One day, in order
Open the Robots report and check that you are not looking for indexing in ordinary statistics. Self-identified robots are not there.
In the working report, turn on “Data: without robots” and compare with data that includes robots. Write down both figures for bounces and for visits. The gap is the behaviour-based layer that is in the report by default.
If you need one slice without that layer, remove it with a segment in this report only. A segment does not rewrite other reports.
Open the Direct summary. Remember that these clicks have already passed fraud protection. Read a bounce as a leave within the threshold after one view, 15 seconds by default.
If the campaign is missing, check the counter link. Unlinked visits go to “Other campaigns”. Look for
yclid, which Direct sets itself.Match a visit under 15 seconds to the three bounce conditions. One view, a short time and no “non-bounce” event make a bounce. Send delay can leave a short visit as a non-bounce.
Read a Webvisor recording as a picture of what the visitor saw, dynamic content included. An iframe will be missing. The recording neither cancels a robotic status nor proves the money.
Network placements do not fit this sequence. Look at them after the Direct summary and the “without robots” mode are already separated.
Reading one visit
This is a teaching example, not a measurement of your counter.
In an ordinary report a visit lasts 9 seconds and has one view. The bounce threshold is still the default, 15 seconds. There is no “non-bounce” event. The visit is a bounce. The same visit is in the Direct summary, so the click passed fraud protection: a bounce in the summary does not move the click back into invalid traffic.
In the with-robots / without-robots comparison the visit exists only in the with-robots mode. Then it belongs to the layer that stays in statistics by default. A segment can remove it from this report. It will not move into the Robots report: that report is for robots that identified themselves.
Webvisor has a recording. The dynamic block of the page is there, and a banner that lived in an iframe is not. An empty rectangle does not prove the banner was absent. The player has no “robots only” button, so you came to the player from a report, not the other way around.
If the visit has no yclid and the summary has no campaign, check “Other campaigns”: the counter may be unlinked.
How to read the pair “with robots” and “without”
The comparison in the report settings is the cheapest way to stop arguing about a percentage. Write down the period, the grouping and both modes. The gap in visits is visits of robots detected from behaviour, which statistics include by default. The gap in bounces shows how far that layer moves the bounce rate, but the help text does not set a “usual” share. Do not paste someone else’s percentage in place of your pair.
A segment does the same subtraction inside one report. It does not create a new source and does not move visits into the Robots report. If the Direct summary does not change after the segment, that is expected: the summary is counted from clicks after fraud protection, not from a “without robots” switch in an arbitrary report.
Keep the bounce threshold separate from both modes. While it stays at the default, a bounce is one view shorter than 15 seconds without a “non-bounce” event. If the counter uses another threshold, the same 15 seconds in a conversation about the Direct summary and in a conversation about the glossary can diverge: the summary speaks of 15 seconds after one view, the glossary of the configured bounce-calculation time. Check the setting before you explain the gap with bots.
yclid on an ad link is Direct’s tag, not a sign of a robot. If it is missing on a visit you think is paid, check “Other campaigns” and the counter link first.
Checklist before you write “these are bots”
Mark yes or no. Any “no” stops the conclusion and sends you back to its layer.
The Robots report is open, and indexing is not being sought in ordinary statistics.
Both modes are written down for the working report: without robots and with robots, one period.
If the layer gets in the way, it is removed with a segment in this report only.
The Direct summary is read as statistics of clicks after fraud protection.
A bounce there is matched to one view and the threshold, 15 seconds by default, plus the absence of a “non-bounce”.
A short visit that did not become a bounce is not declared “a live person” only because of the clock and send delay: the help text allows both outcomes.
A campaign missing from the summary is checked in “Other campaigns”.
An empty iframe in Webvisor is not taken for an empty page.
A robotic visit in the player is not taken for a click that certainly stayed in spend.
Network placements are not mixed into the summary: they have their own report.
Until at least one line is empty, “bots ate the day” is early.
A second example: an unlinked counter
Another teaching example.
The counter shows 400 visits that a marketer thinks of as paid, and the Direct summary has no campaign. The help text puts such visits in “Other campaigns” when the counter is not linked. Until it is linked, comparing the summary’s bounces with CRM leads is pointless: the summary for this campaign is empty for a reason other than antifraud taking the clicks.
In an ordinary report of the same visits, the with-robots mode has more bounces than the without-robots mode. That is the behaviour-based layer, on by default. A segment removes it from the report. Only robots that identified themselves go to the Robots report, and that report is for indexing, not for Direct spend.
One visit in the batch lasts 12 seconds, one view, no “non-bounce”, default threshold. That is a bounce. A neighbour lasts 8 seconds and is not marked as a bounce: the help text allows that because of clock skew and send delay. Neither visit becomes an invalid click from that alone.
Webvisor has a recording of one of them. The dynamic catalogue block is in place, the ad tag in an iframe is empty. That recording is not enough to decide whether the ad should stay on a placement: the player does not filter “robots only”, and a placement block lives in the placements report, not in the visit feed.
Keep the same order the next day. Robots report first, then the pair of modes, then the Direct summary and the counter link, then the three bounce conditions, and only then the player. If you start from the player, an empty iframe and a robotic mark easily become a story about money, while the money in this help text sits somewhere else: the summary counts clicks after fraud protection, and a recording can exist for a visit that is already robotic.
For someone who looks only at bounces, one sentence is enough. A bounce is not a bot label. It is a visit with one view shorter than the threshold and no “non-bounce”, and the default of 15 seconds can be changed. Until that sentence is said, the bounce rate in the summary cannot be added to the Robots report and called click fraud.
What this walkthrough does not glue together
A self-identified robot, a behaviour-based robot, a bounce and a click after antifraud are four answers. High bounces while behaviour-based robots are included are handled first by comparing modes. An empty Direct summary is handled first by linking the counter. A recording without an iframe block is read first as a limit of the player.
You can fold them into “Metrica is full of bots” only after steps 1–8 show the same picture in all four layers. Before that you have four different reports.