Click Fraud: How Buyers Can Recognize That Inventory Is Fraudulent

As you probably know, fraud has become a serious problem for digital advertising. According to some researches it accounts for up to 50% of media advertisement. Recently some players even have started offering money-back guarantees (more details here and here).

So I would like to share some rules which can help Advertisers (and others) to identify that ‘something is going wrong’. Most of these rules are quite straight-forward and simple but our analytics and data scientists have been seeing a lot of examples based on such simple rules:

a) Pure statistical rules which can be checked manually:

  • Abnormally high CTR for a specific piece of inventory (publisher, placement or domain, hereafter referred to as the Inventory). There is no other inventory with such high CTR.
  • A jump in CTR in a specific time period (day/hour) on the Inventory (bucket), which is out of sync with the network.
  • CTR of inventory is heavily increasing while conversion rates stay the same or are even falling (in a look-back period).
  • CTR of <UNKNOWN> element is high.
  • A high number of clicks from “no-cookie” users.
  • The suspected inventory has a very uniform distribution of impressions/clicks between all keywords/geo/browsers/etc. meaning the robot looping identificators and is changing it one by one.
  • The suspected inventory has abnormally high performance from other geo locations.

b) Rules which can be checked based on ad server log-files:

  • A quantity of clicks from the same IP-address.
  • A quantity of clicks from users coming from the same resource.
  • Clicks on all ads displayed at the same time on one page by a user.
  • Use a list of known robots IP-addresses (like that published by the IAB) to check user IPs on the list.
  • Impressions or clicks in equal periods (The behavior above is called ‘periodic’ because the robot generates an HTTP request at a regular interval. A function that looks at the distance between timestamps and calculates the rate of change between them can be tuned to indicate periodicity.
  • A great number of clicks from the same user in a period of time which would be too brief for a human to absorb the impressions delivered. This is detected by defining a window of time, 10 seconds, for example, and counting the number of events (clicks or impressions) occurring inside. If the number is higher than a reasonable threshold, the visitor may safely be declared a robot.

c) Rules based on 3rd party statistics:

  • Big discrepancies between viewability and click rate.
  • A great number of users who exit a landing page in quick time after entering.
  • High bounce rates (exit rates).
  • The user does not move the cursor on a landing page.
  • The user performs no scrolling (90% of robots can’t simulate scrolling).

If you find one of these occurrences, it doesn’t mean that your inventory is fraudulent - it is just a signal that you should pay attention to it. But if you find at least three of these happening then …this should be a trigger for you to take action.

I hope this advice will help you save some money for your business.

Simon Volman
Simon Volman
Chief Technology Officer

Have a look at

Leave Your Contact Details Here

Have questions or would like to see the demo? Get in touch – and we will be delighted to talk