Google and Click Fraud: Behind the Numbers

Last week, we cited a study finding that 16.6 percent of all pay-per-clicks on the Internet were fraudulent in the fourth quarter of 2007, up from 14.2 percent for the same quarter in 2006. The statistic, as reported by MediaPost, was compiled by Click Forensics, an independent auditor that has created the Click Fraud Index. Given what must be billions of total ad-clicks on the Internet, as well as the constant development of click fraud detection programs by online ad providers, the number seemed surprisingly high.

Enter Google, which maintains that its internal click fraud algorithms tell a different story. According to the Google AdWords blog, the company maintains a “3 part system for invalid click detection” that involves filtering out anything funny-looking, analyzing the filtered clicks to determine how many are actually fraudulent, and then investing accordingly. Freakonomics spoke to a Google spokesperson, who explained the discrepancy between the first and second steps in the Google process:

Not all invalid clicks are click fraud — for example, the second click of a double click is not one with malicious intent. But since our goal is to filter out as much click fraud as possible, we cast the net of invalid clicks wide enough to effectively minimize that proportion and ensure that advertisers are only charged for valid clicks. The invalid click rate [i.e., the total number of clicks filtered out by Part 1] has remained in the range of less than 10 percent of all clicks every quarter since we launched AdWords in 2002.

After these figures are analyzed offline, Google maintains that the number of fraudulent ad-clicks not caught by the filters is less than .02 percent of the total clicks on Google ads. So it looks like, on Google’s end anyway, the number of click fraudsters may not be quite so huge, and that the metrics used by third party auditors such as Click Forensics may be flawed. A Google spokesman wouldn’t speculate as to how many people that .02 percent represents (though whatever the number is, it has likely declined, given that the company’s total ad-clicks were reportedly down 7 percent in January). Anyone care to do the math?