Tuesday, August 08, 2006
Shuman Ghosemajumder, Business Product Manager for Trust & Safety, returns today with important information regarding third party detection of click fraud:
A rigorous technical analysis by Google engineers has found fundamental flaws in the work of several click fraud consultants – flaws that help explain why widely quoted estimates of the size of the click fraud problem are exaggerated. We would like to share this research so that advertisers can be aware of these problems and so these consultants can use the information to improve their services.
We provide detailed analysis and explanation of this work here. Two key findings are below, which explain the fundamental flaw we have seen in all of the reports we examined – fictitious clicks: events which are reported as fraudulent, but are never recorded or charged as ad clicks by Google:
- Fictitious ad clicks because of mischaracterizing events. This finding may be the most significant flaw responsible for exaggerated click fraud claims. The problem lies in the fact that many click fraud consultants don’t count actual ad clicks. Rather, to determine the number of ad clicks, they use a number of other signals, including counting visits to a particular webpage. As a result, the consultants count page reloads and subsequent visits on an advertiser’s site as multiple clicks on the advertiser’s Google ad. This generates fictitious ad clicks in the consultant’s reports. For example, if a user browses deeper into an advertiser’s site, then hits the back button, this causes a potential reload of the original landing page, which a consultant would record as an additional ad click – even though no Google ad click actually occurred.
- Fictitious ad clicks due to conflation across advertisers and ad networks. Some consultants “cookie” users and track their activity across their network of client advertisers. One often-used consultant implements the cookie in such a way that clicks on Yahoo ads can be counted as clicks on Google ads, and vice versa.
These kinds of flaws in methodology cause click counts in consultant reports to be artificially inflated. One clear indication that the consultants’ results are flawed: they’re not even getting the total number of clicks correct. We have seen some instances of reports showing 1.5 times the number of clicks in our logs – for example, in one case 1,278 clicks were claimed as being “fraudulent” by the consultant while only 850 actually even appeared as clicks in Google’s logs.
More evidence of the consultants’ defective methodology is revealed when looking at conversion rates. We found clicks identified as “fraudulent” in reports often converted at nearly the same rate (and in some cases better) compared to other clicks. In one case, “fraudulent clicks” converted 5.1% of the time — only a bit less than the advertiser’s overall conversion rate of 5.8%.
Our report provides detailed case studies for three third-party auditing firms – AdWatcher, ClickFacts, and Click Forensics – which represent the vast majority of the last 100 reports advertisers have submitted to us. All of the reports we’ve seen from these consultants exhibit the serious problems we have described above. The pervasiveness of these problems concerns us, especially because advertisers may be hurting their businesses by changing their campaign settings based on erroneous information. We will continue to devote attention to this issue to keep our advertisers well-informed, and perhaps help third-party auditing firms improve their methods so that they can provide value to advertisers.
If you would like to know more about the general issue of invalid clicks and how we manage them, we have posted previously here and here. If you would like to learn more about how to track invalid clicks in your account, you can find information about our Invalid Clicks reporting feature here.