Adsense Click Fraud Estimates by Third-Party Auditors

Google – The big problem with third-party auditors is that they continue to count “fictitious clicks,” clicks that Google doesn’t count as clicks at all, in click fraud estimates.

Here’s one major example: Users click on a Google ad on Google.com or an Adsense site. When they land on the advertiser’s site, they click on products, hitting the “back” button to go back to the landing page. Many browsers reload the landing page each time. We don’t count those as clicks, but third-party auditors actually register each click on the “back” button as another click on an ad, which grossly overestimates the number of ad clicks. 

Counting Clicks

Given that most of Google’s $13 billion in revenue comes from clicks on ads, you would think the words “click fraud” would inspire fear in Shuman Ghosemajumder, the company’s senior product manager and resident click-fraud czar. But the problem–publishers who inflate pay-per-click ad fees with automatic clicking software–doesn’t fluster Ghosemajumder or other Googlers.

In March 2006, as the company faced a class-action lawsuit from online advertisers for damages from fraudulent clicks, Google Chief Executive Eric Schmidt dismissed click fraud as “immaterial.” Even after paying a $90 million settlement, Google has maintained its “everything-is-just-fine” answer to click-fraud worrywarts.

Not all outsiders are reassured. A report in July from Click Forensics, an Austin, Texas-based click-fraud auditing firm, contended that click fraud is growing, particularly on content networks like Google Adsense and Yahoo Publisher Network. The report, drawn from more than 4,000 sites’ advertising data, asserted that the rate of fraudulent clicks was 15.8% in the second quarter of 2007, up 1% from three months earlier. Within online content networks, Click Forensics estimated that more than 25% of all clicks were fraudulent, up from about 22% in the previous quarter.

Ghosemajumder doesn’t buy it. He spoke with Forbes.com about why some third-party auditors overestimate malicious clicks, the myth that Google profits from click fraud and just how the company can reassure advertisers without giving away the secrets of its click-fraud detection.

Forbes.com: Click Forensics reports that click fraud is increasing–you disagree. Why?

Shuman Ghosemajumder: The big problem with third-party auditors is that they continue to count “fictitious clicks,” clicks that Google doesn’t count as clicks at all, in click fraud estimates. Here’s one major example: Users click on a Google ad on Google.com or an Adsense site. When they land on the advertiser’s site, they click on products, hitting the “back” button to go back to the landing page. Many browsers reload the landing page each time. We don’t count those as clicks, but third-party auditors actually register each click on the “back” button as another click on an ad, which grossly overestimates the number of ad clicks.

The other important point is that auditors like Click Forensics estimate the amount of click fraud that’s being attempted, not how much is going undetected and is charged to advertisers. That means they’re counting the clicks that we throw away as invalid, not just the ones advertisers pay for.

But click-fraud auditors argue that there’s a discrepancy: They say you throw away no more than 10% of clicks, while they estimate click fraud rates at as much as 25%.

That’s just one particular set of numbers. The auditing firm, Fair Isaac, for example, estimated in May that on Google’s content network, 10 to 15% of clicks are fraudulent. On ads placed next to search results, they said that there was a negligible rate of click fraud, less than 1%. That implies an overall click-fraud rate of around five to 7%. The number of clicks that we proactively throw out is less than 10%. So then the question is really: How much are advertisers getting for free thanks to our detection methods?



Most auditors have realized this is very difficult to do from the point of view of a third party and have changed their business model to focus on other kinds of analytics. Click Forensics is one company that has continued to publish numbers, but I don’t hear much about them anymore. You need a lot of data to do any sort of statistical analysis, and auditors don’t have nearly as much as we do.

What data do you see that third-party auditors don’t?

The primary data set that third parties don’t have access to is the impression data–not just when the ads are clicked, but how often they’re viewed. A very simplistic fraudster might just click on ads, over and over. But of course, we’ve learned to recognize spikes in the click-through rate, the number of times the ad is clicked on for every time it appears. Any sophisticated fraudster has to keep their click-through rates at a reasonable percentage by creating the right number of false impressions.

From a third-party perspective, you can’t even see the click-through rate. Since we know exactly how many times an ad appears, we have much more data than the third-party auditors for any given advertiser.

Have you seen an increase in click fraud in recent months?

It’s difficult to say. We filter out fewer than 10% of clicks as invalid, but you can’t draw a very precise line between malicious activity and invalid clicks that might have been made with malicious intent. Anyone spending time looking at the percentage changes in something where you can’t draw a precise line doesn’t understand the issue.

One metric we do have is the number of cases in which we refund advertisers for click fraud, and that has remained stable at less than .02%, quarter after quarter.

Some have accused Google of turning a blind eye to click fraud or underestimating the problem because of the revenue it generates. Is Google really motivated to stop click fraud?

That’s one of the biggest misunderstandings about this. We have a very direct financial incentive to detect bad publishers–the ones engaged in click fraud–and remove them from the system.

Google is also a publisher itself. When we display ads on Google.com, we get 100% of the revenue for those ads. When you have bad publishers introduced into the system, the effect is to reduce advertisers’ return on investment, because they cost money and don’t generate sales. When you reduce [return on investment (ROI)], the rational thing for advertisers to do is to pay less for clicks. That doesn’t just mean that the clicks that go through our good publishers’ sites make less money. It also means the clicks on Google.com make less money for us.

That incentive for us to stop click fraud is actually part of a broader incentive: We’re trying to create the greatest ROI possible for advertisers. That’s what makes us the most competitive in terms of attracting advertisers and competitors. We wouldn’t invest a lot of money improving ROI in one respect, and then let it degrade in another.

Has Google’s approach to click fraud changed since its $90 million settlement with advertisers last year?

We weren’t required by the settlement terms to change anything. But we did have a third-party analyst, Alexander Tuzhilin, come in and spend time with the click-quality team. He wrote a report that concluded our approach was reasonable. We’ve felt we’ve done a very effective job of protecting advertisers, even though we have fraudsters constantly trying new angles of attack.

Are you investigating and taking legal action against particularly egregious cases of click fraud?

We are trying to do that more, and we’re looking at a few cases at the moment. But I’m afraid I can’t talk about them.

What are the giveaways that make click-fraud detectable?

There aren’t any silver bullets. There are some very simple things, like watching the click-through rate. It’s easy to catch someone who clicks a million times in a minute. But we also have anomaly detection that allows us to spot more sophisticated kinds of click fraud.

We monitor signals like Internet protocol addresses and the kinds of browsers people use–for traffic to look real, it has to have the right proportion of visitors using the Firefox browser, Internet Explorer, et. cetera. But the real power of our detection systems lies in the fact that we analyze hundreds of different factors, the majority of which are secret.

Does that mean that Google faces a transparency dilemma? You can’t reveal these metrics because they’d help people get away with click fraud. But if you reveal too little, advertisers will have a hard time believing that you’re really capable of stopping the problem.

We can’t talk about these things in detail, but we talk about them at a high level. We try to promote as much understanding of these issues as we can in forums, blogs and academic articles. And we’re the only search engine to have had a third-party come in and publish a report.

What advertisers can see and what they’re constantly giving us feedback about are the actual results from their ad campaigns. The metric we really compete on is our advertisers’ return on investment.

Some critics have also accused Google of unfairly banning innocent publishers accused of click fraud. Does that happen?

We do take it very seriously. We have a zero-tolerance policy when it comes to click fraud from publishers.

But when it comes to accounts that have been wrongly terminated or sabotaged, we’re always monitoring that too, and it generally comes to a very low number. We never hear back from the vast majority of publishers terminated. When publishers say they weren’t engaged in click fraud, we have an appeals process.

But more generally, it’s a good sign that we’re being criticized by both advertisers and publishers for opposite reasons. That means we must be doing something right.