Bounce rate: what it says about traffic

Bounce rate in Yandex Metrica answers a narrow question: did the visit leave after a single page view, before the threshold, and without the service event “non-bounce”? The glossary requires all three conditions at once. Miss one, and the visit is not a bounce. This is not a fraud test and not a grade for the ad.

How robots look in the same counter is covered in how to find bots in Yandex Metrica. Below is what actually gets marked as a bounce, where the threshold changes, and why a high percent still says nothing by itself about click fraud.

Three conditions, all required

A visit is a bounce only when three conditions hold together: no more than one page view, a duration shorter than the bounce calculation time, and no service event “non-bounce”. The default time is 15 seconds. It can be replaced. It is not a fraud threshold.

The browser sets the “non-bounce” event after that time. Because of clock differences and send delay, a visit shorter than 15 seconds may still not be a bounce. A person closed the tab at the tenth second, the packet left later, and the event was recorded — the report no longer calls it a bounce. The reverse sits in the same rule: one page view longer than the threshold is not a bounce, even if there was no second page.

A worked example, not a site cut. Of 200 visits, 50 meet all three conditions. Bounce rate in this example is 50 / 200, or 25%. Another 20 visits lasted 10 seconds but already have the “non-bounce” event. They stay out of the numerator. Treating “every short visit” as a bounce adds people the glossary has already taken out.

Where the 15 seconds are set

The counter initialization parameter accurateTrackBounce controls the accurate bounce rate. true turns it on and sets the “non-bounce” event after 15,000 ms. false turns the accurate rate off. An integer sets the same delay in milliseconds. The main initialization parameters are assembled on the Settings page.

Fifteen seconds is the default value of the parameter, not a bot threshold. The three glossary conditions do not go away with the parameter. Only the duration the visit is compared with changes. Setting 30,000 ms moves the second condition. It does not teach the counter to tell a robot apart. Setting false turns the accurate rate off. It does not zero bounces in every stored report: the help page describes initialization, not a rewrite of the archive.

In the sources summary Metrica says it directly. The higher the bounce rate, the less the site engages visitors: they leave within 15 seconds after one page. A high bounce rate on visits from search is named there as a possible sign that the pages are not relevant to the queries. That is a reading of engagement and relevance. The 15 seconds here are the same default as in the glossary. The help page does not suggest deciding from this percent that the visits were bought.

Bounces in the Direct report

The “Direct, summary” report opens at “Reports → Sources → Direct, summary”. Its statistics are based on clicks that passed fraud protection in Direct. The help page ties bounces in this report to whether the page was left within 15 seconds after one view. If the counter is not linked to the campaign, the visits fall into the source “Other people’s campaigns”. Direct itself adds yclid to ad links.

The summary is therefore not about every click in the account. Clicks that Direct’s protection did not pass are not the base of this report. Comparing its bounce percent with a report of all site visits compares different entrances. In both places, 15 seconds is the default threshold until you replace it on the counter.

Robots in the numerator

By default, Metrica statistics include robot visits detected by behavior. The help page gives the example outright: robots can lower page depth and raise bounces. The page does not give a share they “usually” add. A rise in bounces fits an empty landing page, robots in the sample, or both at once. The percent does not sign the cause.

You can compare a cut without robots and with robots in the report setting “Data: without robots”, with the action “Compare with data including robots”. A segment can take robots out of one report. That is a comparison of one report, not a correction of campaign history. What Metrica calls robotness, and how far that label can be trusted, is in robotness in Metrica.

A bounce in a retargeting segment is a different rule

Direct’s glossary calls retargeting a showing to people who have already been on the promoted site. Among the automatic segments there are ready-made conditions.

The bounce period can be narrowed in the segment condition. The segment’s definition of a bounce does not repeat the “non-bounce” service event from the Metrica glossary. A short visit that has the event may not be a bounce in a Metrica report and may still fall in or out of the segment under a different wording. The page does not say robots are removed from the segment. Showing ads to the “Bounces” segment is a showing to people who match that condition, not a fraud filter.