Andy Beal over at MarketingPilgrim has a nice article on how Google’s click fraud rate is less than two percent, at least according to Google’s own figures. This is based off the "invalid" click rate that Google gave him, which includes click fraud and clicks that aren’t actually billed. I could have sworn that Google had given this type of invalid rate before but can’t track that down — so that definitely seems new. The idea that Google self-reports a click fraud rate far below that of some third party estimates is definitely not. Here’s a dive into Andy’s article, along with some other reports for context.
Andy talked with Google’s Shuman Ghosemajumder, business product manager for trust & safety, about click fraud detection. Shuman explained that a chunk of all clicks are considered "invalid" and not charged to advertisers. Indeed, back in July, Google made it possible (and also see here) for advertisers to see exactly the percentage of invalid clicks that Google was screening.
Not all invalid clicks are fraudulent clicks. For example, someone "double clicking" quickly on an ad would cause the second click to be deemed invalid, and the advertiser wouldn’t (or shouldn’t) be billed (see About invalid clicks from Google for more on this). But some of those invalid clicks are fraudulent, people trying to make money by clicking on things like their own AdSense ads or trying to hurt competitors by clicking on their links.
A portion of fraudulent clicks are NOT caught by Google’s filtering. That’s a known fact, given that Google itself issues refunds. What’s the percentage? Google doesn’t say. Plenty have asked, such as when I asked Google CEO Eric Schmidt about it earlier this year:
Danny: Google must know the rate of click fraud that is escaping [you], or have some real estimates, but we never see numbers. It’s fair to say Yahoo! doesn’t give me a number either. Yahoo! will say "we’re better than industry standards." And I’ll say "what are you at?" [And they say] "Well, we’re better than industry standards."
Eric: Just so you know, last week we announced a program where our advertisers can see our estimate of the click fraud rate. So that’s an example, and hopefully a good response to your question. To the degree that our systems can detect click fraud, which is a very serious problem, we detect it. We believe we have it under control. We believe that that is getting better. And so we’re trying to get advertisers more of a sense of how it’s working within the limits of how our systems work.
Danny: That’s a great example, because on the one hand, that is more transparent— but then what you don’t tell us is, here’s the total amount of money that was spent, here’s the total amount of money we issued in refunds for click fraud that got past our system. You do the math, that’s your rate.
Google has said click fraud is in the single digits on average, so at worst, less than 10 percent. That brings us to Andy’s post today. He was shown a chart listing the various ways that invalid clicks are filtered, with bars in proportion to the amount of invalid clicks detected. Looking at the small bar of "requested investigations" — the percentage of invalid clicks investigated because Google itself has not flagged them — he asked if this was in the one to two percent range. Shuman wouldn’t give him a figure but did say Andy’s estimate was either near those amounts or that they were less.
So to be clear, two percent of all clicks are investigated by Google as being possibly fraudulent after billing has happened. Some amount of fraudulent clicks, of course, are never reported. That will take the figure up. However, some amount of the investigated clicks will be cleared of fraud, taking the amount down.
Of course, instead of this guesswork, Google could really be transparent and do exactly as suggested above — report how much money it pays out in refunds for clicks deemed to be fraudulent. That would form the basis of a real, self-reported rate from Google. By the way, Yahoo could do the same.
Yahoo made much play when announcing its own clickfraud settlement that it was found to have a clickfraud rate below the "industry estimates." These are estimates that I’m fairly certain Yahoo had previously said were inflated. Google certainly has said that. I asked Yahoo if they were found to be so good, why not just publish their actual clickfraud rate. Like Google, they declined.
For more on the internals of click fraud at Google, see my past article The Abridged Version: Independent Report On Google’s Click Fraud Detection Practices, which has an expert witness in the click fraud case that Google settled examining the click fraud detection system over time, 47 pages of examination!
For the issue of why Google disagrees with higher estimates third party firms have given, you can see this post Google made back in August that took aim at some companies. Fair to say, some of those companies don’t agree with Google. A recent estimate came out in October that prompted Google to state again that the reports were "considerably" more than it sees internally.
Postscript: See also Revisiting Google’s 2 Percent Or Less Click Fraud Rate