Google says click fraud is less than .02%

Google reports today that click fraud is less than .02% of all ad clicks. Danny Sullivan at SearchEngineLand has a detailed analysis of the Google claims. I have long suggested that click fraud is lower than most estimates, but these numbers from Google surprised me.

Click fraud involves people or bots clicking on ads to generate revenue for themselves or to hurt competitors by using up their ad budget. For example, lets say you own a web site about cars. You make a deal with Google to provide advertisements on your site. When a visitor clicks on an ad on your site, Google collects revenue from the advertiser, and pays you 40% to 60% of the revenue. What could be better? Well, some unscrupulous web site owners figured out that if they had their employees and friends click on the ads it was a revenue machine to them. Then they figured out they could write some software (bot) to automatically click on the ads at random times of the day. Ka-ching! More revenue.

Google, Yahoo, and Microsoft got wise to this problem very quickly. They put filters in place to catch invalid clicks. The click fraud filters have all sorts of rules built into the algorithms to define and detect invalid clicks. For example, if a user clicks on an ad twice in a second one of those clicks is likely invalid. Or, if the same IP address clicks on the same ad 10 times in an hour it is probably invalid.

The click fraudsters developed more clever methods. They knew that Google was watching so they started to randomize their bot clicks to less than 1 per second, made multiple clicks at random times of the day, used hundreds or thousands of different IP addresses, and all sorts of tricks to stay "under the radar" and avoid detection.

The search engines respond with even better detection filters. This is a cat and mouse game similar to email spam detection or encryption techniques. The search engines can't reveal their secrets because the fraudsters will adjust their click fraud attacks to avoid detection. Lets just say there are many ways to detect what machines and IP addresses are clicking on ads, and there are many models of normal click behavior that can detect unusual behavior.

Google says that less than 10% of all clicks are invalid. Of this 10%, Google says they detect and filter out 99% and the advertiser is never charged. What remains is .02% of all clicks that might be fraud. Google will provide refunds to advertisers who can reasonably prove fraud.

What about the advertising network sites? I believe the search engines can indeed detect click fraud for traffic on their own search sites. But, it gets much more difficult for clicks that originate on the millions of sites in the advertising network. It can be done, but it is difficult because each site has a different "normal behavior model" meaning, normal traffic and clicking patterns. But, like anything else, it is just a matter of scale. If the algorithm is right and self adjusting, it can be applied to any site and detect unusual behavior.

It is all about ROI on your advertising dollar. In the end, whatever the click fraud rate, it is all about the Return On Investment for each advertising dollar spent on search engines versus the ROI on TV, radio, newspaper, magazine, or other advertising methods. At least web advertising has real trackable numbers for response rates and some targeting on where the responses came from. Not so with TV, radio, and magazine ads.

Web advertising is still less than 10% of all advertising budgets. I think we will see more and more advertising budgets moving to web advertising. Now that the click fraud problem is "contained" to a reasonable level more advertisers will feel comfortable moving more budget to the web.

Subscribe - To get an automatic feed of all future posts subscribe here, or to receive them via email go here and enter your email address in the box in the right column.

Comments

[Marcelo Calbucci]

Hi Don, this data is interesting because it shows a big difference to what Google tell us and what we are measuring. See my post on the subject:

http://marcelo.sampasite.com/brave-tech-world/e/Google-Click-Fraud-is-at-odds-wi.htm

[CPCcurmudgeon]

As a system that provides best effort services to identify some noise in click patterns, I don't have a problem with this. OTOH, they haven't proven (rigorously and scientifically) that it actually detects the amount of click fraud they say it does. Especially problematical is the fact that what is considered fraudulent is mainly a matter of what an advertiser thinks is fraudulent. If the advertiser has a big budget, and can absorb a lot of fraudulent clicks, the system may not catch them until the advertiser complains.

All this aside, I still wonder why it has taken so long for them to do what they should have done in the very beginning when AdWords CPC was launched. The system under discussion now should have been what was designed before AdWords CPC went live.

I would still encourage advertisers to watch their budgets carefully, bid cautiously, and scale back spending at the first sign of unusually nonconverting traffic.

[Joe Duck]

Don an excellent summary of the cat and mousing but it's *not* just about ROI. Google, MSN, Yahoo have an obligation to find fraud. This study is encouraging as is Shumander who I think is a very honest guy, though I remain skeptical they are catching, for example, all the pay to click schemes. They use quality algos to determine "worthless" clicks but it seems clever scammers would buy "a few" items to get past these. It's not clear to me this can be measured or if it even falls in their definition of fraud.

[Steve Morsa]

All in all, this long-overdue "proof/disclosure" by Google can only help and support all the players and participants in the paid search/PPC industry by showing that click fraud is--apparently--not as big a problem as has been generally believed...

...and really, given that no real "secrets" were revealed in doing so, it was silly to not "reveal" this information to everyone years ago.

I agree with you, Don, that ROI is--and should always remain--the most important consideration in the use of PPC. With today's sophisticated result measuring tools, as weird as this sounds, even if click fraud was 90% of all clicks, if the ROI was there, advertisers would still be wise to use it...though of course it's the right and just thing for the providers to do all that they can to prevent advertisers from having to pay for such fraud...

...of course, at least manual click fraud could crippled, if there was some way for advertisers to deliver their PPC ads only to exactly and precisely targeted recipients, instead of with easily ID'd and accessed keywords.

Now all we have to do is agree on just what "click fraud" is...

[Erik Schwartz]

Wouldn't this really mean that Google detects .02% click fraud? To be successful click fraud must not be detected.

[Bernard Gallet]

Don, I do not believe that Click Fraud is so small. For example, for our clients, the average origin is 64% in the US, however, on the content network, it goes down to 34%.

Also, our tracking uses some javascript, does not is look odd that the proportion of browser without javascript is much higher in the content network?

And I also do not agree that ROI is the only important measure. Let's say you sell something on your site for $100, with a COGS of $60. If you pay $30 per action, you end up with a profit of $10, so this looks good. However, assuming a 10% click fraud rate, as reported by a lot of companies (not Google :-)), you could have paid only $27 per action, hence increasing your profit by 30%...

The only way for Google to prove their claim is to be a lot more trasparent than they are today.

[Bob]

Meanwhile, we get this:

Google says it loses $1 billion a year to false ad clicks

http://www.cbc.ca/technology/story/2007/03/02/tech-googleclickfraud-20070302.html

This talks about something ~10% as a fairly constant "invalid" click rate. So I guess we're meant to believe they catch almost all of that? LOL. I think it's highly unlikely that GOOG knows exactly what the extent of click fraud is and especially within their partner-generated traffic. I also find it hard to believe they're super motivated to figure it out given that, despite what this says, advertisers are almost surely getting billed for some of it and ALL that flows to GOOG.