The switch "Filter robots by behavior" deletes those robots from source data, and it has no undo. How to compare a report with a segment, and when the help says to turn the filter on.
Ten steps to separate invalid traffic from live people with no interest, and from a strategy that does not yet have ten goal achievements a week.
How to separate self-identified robots, behaviour-based robots, bounces and a Webvisor recording, and which Direct report already counts clicks that passed antifraud.
One day in Yandex Direct and Metrica: a suspicious Webvisor, what counts as a visit, and how to block a placement from the Placements report. A checklist without extra accounts.
The checked guides never name a season as the cause of a conversion drop. A drop you can see comes from a different report date, from campaign types you must not compare, or from a bounce threshold you can change.
A goal works when Metrica counts the action you named. Fewer than 10 achievements a week is the strategy’s learning line, and Direct will not remove a second view of the thank-you page.
The checked guides have no “one device, many visits” report. A repeat click on an ad inside a visit that is still open does not become a new visit. The default timeout is 30 minutes.
The checked Direct guides do not describe an hourly click report. The time grain they do name is one day down to minutes, and a sharp rise in clicks can happen with no invalid traffic.
Do not put search and network conversion in one column. Display campaigns on search are not compared with performance campaigns, and two Direct reports date the same conversion on different days.
In the Direct report wizard, “where the click came from” is four groupings and the Clicks metric. You can block a placement in that table, and the guide never turns “significant data” into a click count.
Do not add impressions, clicks and conversions in this report into one loss. The number can still grow, and Metrica’s Direct summary shows the clicks the filter has already let through.
Clicks without leads are a choice of order. First what is already off the bill, then the goal, the page and the offer. The click-fraud branch opens when those checks are closed and the click stayed in spend.
In the visit dimension the last byte of IPv4 is zeroed. A counter filter on an address does not show the past and irreversibly narrows the future. These pages do not name a provider.
The Placements report already shows the name, spend, clicks and conversions on networks. One or two conversions are not enough for a conclusion, and the guide calls manual disabling unsound.
Webvisor records robot visits too, and the player does not stamp them as bots. An idle mark, an iframe hole and asterisks in fields are not a verdict and not spend.
Open Site load first, then Robots for the spike. A row counts data sends, not visits and not Direct spend.
A hundred clicks and seventy visits are not a missing row. Direct counts a display click before the site, a repeat inside a visit is not a new visit, and the Direct group counts only clicks that passed antifraud.
A higher click price on the same budget is not a fraud column. Direct’s predicted bid already includes the CTR forecast and post-click behavior, and invalid clicks stay out of spend.
Clicks with no leads come from a broken goal, a real person who did not buy, a click that never reached the site, or an empty click left in spend. Check in that order. Fraud is last.
A sign of click fraud is a click that stayed in spend and brought no target action, after Direct's filter left it in. A spike in Webvisor is not that sign by itself.