Combatting Click Fraud

September 1, 2005 (DRAFT) — I’m going to share some strategies for “effective pay-per-click search marketing,” including strategies for combatting “click fraud.”

As marketers, we want a positive return on investment (ROI). Thus, if my client spends $1,000 on a “pay-per-click” (PPC) search campaign, she will want to generate more than $1,000 in gross profit as a result. In most cases, a campaign won’t be sustained if the ROI is not positive.

Let’s start by defining some key terms, and identifying some of the key players in this space:

• “Pay-Per-Click” or PPC: An arrangement in which an advertiser agrees to pay a certain amount of money for each consumer who is sent to the advertiser’s web site. A contract between the parties defines the exact situations in which compensation is due; for example, it may restrict the ways a consumer may be encouraged to click on an advertiser’s link, and it may count only the first “click-through” from a specific consumer.
• “PPC Search” is the term I will use to refer to a PPC arrangement in which the primary source of consumer clicks is through a general-interest “search engine” such as Google or Yahoo.
• Yahoo Search Marketing (YSM) is the new name for Overture.com, which was earlier known as GoTo.com; this was the dominant company in this space in 1999-2002 but is now a distant second to Google. Although Overture/GoTo offered its own search engine for consumers to use, most of its revenue came from ads placed through hundreds of other search engines.
• Google AdWords is an automated advertising system in which advertisers may create text ads to appear alongside the search results for specific search phrases, on Google and partner search sites.
• Google AdSense and Yahoo Publisher Network: These are essentially advertising networks which allow AdWords text advertisements to appear on “content” web sites; these ads are not associated with a specific search phrase used by a consumer, but instead Google’s technology attempts to match ads to the site based on keyword analysis.
• Other PPC Search Companies: In addition to YSM and AdWords, there are dozens of smaller companies which use the same general business model.
• Click Fraud: Any activities which are designed to either (a) draw income from PPC search by creating the false appearance of an end-user search and click; or (b) deplete an advertiser’s account through improper clicks, usually by a competitor.
• “Idiot Click”: I define this unique term to refer to a consumer’s action in clicking on a paid advertisement simply because it is the first result shown for a particular search term, without actually reading the text of the ad to determine if it is relevant to the search.

Any discussion of “effective PPC marketing” or “click fraud” can quickly become incomprehensible, because of all the industry buzzwords and variations, so I’m going to just choose some arbitrary examples. Accounting concepts and terms may also be new to you, so I’m going to simplify everything as much as I can. For this discussion, let’s assume that my company (All Widgets Inc.) sells widgets online, through a web site called AllWidgets.com; my largest competitor is “Best Widgets” (BestWidgets.com). Let’s also assume that I sell widgets for $10 each, and that my cost for each widget is $3, plus I estimate that I’ll have general “overhead” expenses that equal about $2 per widget. Thus, if someone comes to AllWidgets.com and buys a widget, I’ll earn $5 “gross profit.”

Currently, without any paid advertising, our web site is currently drawing 5,000 visitors each day, and draws 100 orders per day; the average order is for 2 widgets.

My job, as the Search Marketing Manager for AllWidgets.com, is to draw new customers to the AllWidgets web site to buy widgets. If I can profitably draw a lot of new sales, I will earn bonuses or commissions; if I don’t, I’ll be fired in disgrace. (One more thing: my boss says that my salary of $3,000 per month, plus my bonuses and commissions, will be counted as an expense against gross profit, so I am starting with a “handicap.”)

Widgets – buy @ AllWidgets.com Low prices, best quality, same-day shipping; 100% guaranteed. www.AllWidgets.com/

Starting Off: Widgets are cool, and people want them, and of course lots of people go to Google.com and search for a place to learn about and buy widgets. So I’m going to start by signing up for Google’s AdWords program, and creating a text ad to sell widgets. I’ve decided that if someone searches for “widget” or “widgets” on Google.com, I want my text ad to appear; a sample ad is shown at right. So I fill in Google’s forms to request that my text ad appear any time someone searches for “widget” or “widgets.”

Danger: The first mistake I might make is to accept Google’s “recommendation” for a maximum bid amount for my search term. For example, Google might recommend that I choose “$1.18” as my maximum bid, because there are other bidders who’ve chosen maximum bids of more than $1 per click.

Sorry, Google, but this “recommendation” doesn’t pass the “smell test” for me. I already know that my site draws 5,000 visitors per day, but only 100 orders, so only 1 of every 50 visitors will place an order. I also know that my average order is for 2 widgets, so my gross profit per order is only $10. If my gross profit from 50 visitors is $10, then I should spend no more than 20 cents per click (.20 x 50 = $10).

Since I don’t know whether Google’s users will buy at the same rates as my current customers, I’ll try a maximum bid of just 11 cents per click.

Another Mistake: Did you notice my other mistake? I only bid on two search words: “widget” and “widgets.” But there are many other search words and phrases that may bring more qualified people to my site. For example, how about “buy widget” or “bulk widgets” or “case lot of widgets”? If people use widgets to repair their fromitzes, then I should consider bidding on “fromitz” or “repair fromitz” as well. There are probably thousands of potential search terms to consider.

Anyway, let’s keep going: let’s run the campaign for a while. Google lets me set a daily “budget” for my campaigns, and usually recommends an amount; Google’s traffic estimator also tells me how often people search for this phrase, and let’s assume that there are 100,000 searches per month for “widget” and “widgets.” I’ll arbitrarily choose a budget of $50 per day.

Tracking Code: Oh, yes, one other very important thing: I ask our company’s webmaster to implement a “tracking code” so that we can trace the source of each new order. Thus, my text ad doesn’t just link to the home page, but also includes a tracking code, like this: http://www.AllWidgets.com/index.htm?source=AdWords+widget. Then, if someone clicks from Google to my site, and then places an order, that tracking code (“Adwords+widget”) will be associated with the order.

So there we go. I launch this campaign, and the very next day I’m pleased to discover that 100 people clicked on my ad, at an average cost of seven cents per click (total cost: $7 for the first day). I run a sales report, and discover that three people placed orders, for a total of 6 widgets. My gross profit was $30 but my cost was only $7, so I’d call that a success.

Beyond Google AdWords: Now it’s time to expand my campaign: maybe I could increase the maximum bid amount, or add more keyword phrases, or maybe it’s time to expand my campaign to other search engines.

Gosh, this PPC search thing is amazing. I go ahead and sign up for “another PPC search engine.” I don’t want to single any company out, so I’ll call the company “GleepSearch.com” (which is an unregistered domain as of August 2, 2005).

I’m feeling confident (maybe a little bit cocky), I use the exact same text ad, and the same same bid amount, and then I go home for the weekend.

On Monday morning, I check the results. On Google AdWords, I’ve spent a total of $35, which drove sales of 18 widgets. On GleepSearch.com, I’ve spent $28, but no sales.

For some reason, I do nothing else, and two weeks later, I’ve spent $380 on AdWords to sell 155 widgets, but I’ve spent $350 on GleepSearch.com to sell just 5 widgets.

The final result of the first two weeks of PPC search marketing: I’ve spent $730 to drive sales of 160 widgets, which yield a gross profit of $800. Maybe I won’t get fired (yet), but these are not promising results.

How to Improve Results: The obvious thing to do is to end the campaign at GleepSearch.com, and stick with Google AdWords. That will cut expenses almost in half, while losing only a few sales. But if I stop there, I’m probably never going to find a way to justify my salary.

But I can’t fix anything unless I know what’s broken. So I ask my company’s technical team to find some data for me, and I probably request access to the web server “log files,” and I use a web log analysis program to generate some reports.

I make an amazing discovery: I specified only two search phrases (“widget” and “widgets”) but my web logs show that Google and GleepSearch.com have both sent me many people who searched for longer phrases, like “widget repair” and “free widgets.” When I look to see who placed orders, I discover that most people who actually bought widgets from AllWidgets did not search for “widget” or “widgets.” Instead, they searched for “cheap widgets,” “widgets fast,” “fromitz widgets,” and one customer had actually searched for “Best Widgets.” (Recall that my biggest competitor is BestWidgets.com).

Suspicious & Strange Data: As I review my log analysis results (perhaps from WebTrends or another program), something else confuses me: when people came to AllWidgets.com from Google, they viewed an average of 1.6 web pages and 5.6 graphics files per visit (there are 4 images on the home page), but people who came from GleepSearch.com viewed an average of 1.1 web pages and 3.3 graphics files per visit. In fact, it appears that there are hundreds of visitors who clicked from GleepSearch.com and loaded the main HTML home page, but their web browsers never “requested” any of the graphics on that page!

I also notice that the “referrer” fields for many of these paid searches show a different web site, other than Google.com or GleepSearch.com. There are searches from AOL and other search sites that partner with Google, but there are also search engines I’ve never heard of. I also discover that the “referrer” field is blank much more often for paid search traffic than for other sources of traffic to my web site, and that GleepSearch has many more “blank referrers” than Google.

Something else is odd: my web site’s traffic is strongest during the business day, so that my reports have a “curve” that rises from 8 am to 1pm and then slides down from 2pm to 7pm, with almost no traffic from 1 to 3 am. But when I view the traffic from GleepSearch.com, the curve is much “softer” and “flatter,” and nearly all the traffic between 1 and 3 am is coming from that one source.

Finally, I notice another strange phenomenon: the same strange searches seem to be repeated on GleepSearch.com multiple times. For example, the search phrases “repiar widgets” and “widget frobishko” both appear three times each in referrer strings from GleepSearch.com, but never from any other source.

While I’m trying to figure out what’s wrong with my campaigns, I search on Google and elsewhere to find answers, and immediately I discover a phrase called “click fraud,” and there are claims on several web sites that GleepSearch.com is plagued by fraudulent activity. When I call the company, they acknowledge that in the past, some criminals have tried to abuse their partners by generating fraudulent traffic, but the company has complex systems in place to detect fraud, and my account traffic is unaffected.

---

What Can I Do? In the examples above, I’ve tried to identify some of the most common “inefficiences” that I’ve uncovered while analyzing PPC search campaigns. Now, I’ll offer some strategies to combat both fraud and inefficiencies in PPC search campaigns.

1. Choose Keywords and Phrases Carefully:
   • More Specific: A more specific search phrase is usually more likely to result in a sale. Thus, people who come to your web site after searching for “buy widgets” are much more likely to buy your widgets, than people who searched for “widgets.”
   • Avoid “Bad Searches”: The most obvious “bad keyword” is “free.” People who search for “free widgets” are probably not planning to spend money on widgets. Words like “cheap” or “discount” are qualifiers that may be wonderful for discount resellers, but awful for other merchants.
   • Change Your Bid Amounts and Tracking Codes: You will find that some search phrases are worth more than others. In most cases, you won’t want to spend any amount on “free widgets” but you might be willing to pay a huge premium to present your ad to people who search for “widgets next day.” You can learn that faster by using a unique tracking code for each search phrase. Let me repeat that: use a unique tracking code for each search phrase!
   • Echo the Search Phrase: You should also use a unique ad for each small group of search phrases; your ad headline and text should “echo” the search phrase. Not only does this improve clickthroughs, it will reduce the “minimum bid amount” demanded by Google’s AdWords system (this is a new effect of the changes Google made to AdWords in late August 2005).
2. Watch for Pattern Variations: Inefficient (and fraudulent) search traffic can often be quickly detected by looking for “different behavior.”
   • Abandonment: Every web site has visitors who come to the home page, but then “abandon” the site to go somewhere else. Watch for significant variations in this rate by source.
   • Path Analysis: On most web sites, there are “busy” and “unpopular” sections. For example, very few people read a web site’s “privacy policy” or “about us” page, but many more people will click to view “closeouts” and “today’s special.” Most visitors view only one or a few pages at your site (hence an average visit of 1.7 pages is not bad), and very few view 10 or more pages during a session. If you discover that visitors from a particular source have a dramatically different pattern of activity, you need to pay attention.
   • Double-Click Activity: It is fairly common for a consumer to search for something at Google, visit an advertiser’s web site, and then search again at Google and click again to the same advertiser’s site a few minutes later. This is a normal pattern of activity for someone who is “shopping around” or comparing prices, and it’s not unreasonable to pay two “per-click” fees for this. However, if two clicks come from the same visitor within a few seconds, there should be no charge for the second click, nor should there be any charge for a third or subsequent click.
   • “Echo” Activity: If you discover a consistent pattern in which the same search phrases are executed more often than “normal,” or in which multiple users follow the same unusual path through your site, it is possible that the “echoed” activity is from an automated “agent.” (This is one use for data from “spyware,” which installs itself uninvited on consumers’ computers, and records their keystrokes or web visits; the data can be sent to another computer where the same searches and clicks are repeated.)
   • Phrase Surges: Unsophisticated “click fraud” operations will flood a search engine with many requests for the same search term. For example, I might discover that normally, 10% of my traffic from search engines is associated with a search for “fromitz,” but then suddenly there is a huge surge in traffic through GleepSearch.com, of which 90% are searches for “fromitz.” Sometimes there are valid reasons — maybe Jay Leno made a joke about fromitzes and widgets — but often this is a sign of fraud.
   • Missing Data: Often, an increased rate of “missing data” (such as user-agent or referrer data) may reflect fraudulent activity.
   • Excess Foreign Traffic: In one case where I uncovered “click fraud, more than one-third were from IP addresses in China, and 56% were from IP addresses in non-English-speaking countries. Only 44% of traffic came from English-speaking countries. This is an extraordinary pattern, when compared to typical traffic distributions.
3. Demand Genuine Cooperation from the Search Partner: Google’s “AdWords” staff and the staff of other PPC search systems have a duty to pro-actively monitor search activity and flag suspicious traffic. In most cases, they should delete fraudulent activity from all client accounts without any request. However, it turns out that some PPC companies are more honest than others.
   • Several years ago, I discovered fraudulent activity by “partners of a partner of a PPC search company.” In other words, I paid company “E” for PPC search traffic, and they in turn paid company “F,” which operated its own search engine, and company “F” in turn paid company “G” and dozens of other small search engines for traffic. I discovered that company “G” was generating immense volumes of fraudulent traffic, and so I called Company “E” to ask them to take action. Eventually, they disclosed that company “F” was involved. When I spoke with company “F”, they told me that they had discovered the fraud, cancelled their contract with company “G”, and credited all activity back to company “E” — in other words, they had already told company “E” that the traffic was fraudulent and paid them back for it. However, company “E” refused to reimburse me for this traffic, claiming that there was no proof that it was fraudulent — even after I gave them log files showing the inexplicable activity.
   • Work with legitimate PPC companies to analyze fraud: When dealing with a PPC company like Google AdWords or Yahoo Internet Marketing (Overture), before complaining to the company, you should check to see if you were even charged for the “fraudulent” activity. For example, you may discover 50 legitimate and 500 fraudulent clickthroughs to your server, but Google charged you for 50 clicks, so they have already detected and erased the fraudulent charges from your account. If Google has charged you for the fraudulent activity, be prepared to send them log files for all the activity in question, and insist that the data be given to someone in their “fraud detection” team. (You do not need to hand over your complete log files, which contain valuable proprietary data — I generally use Microsoft Excel or Access to extract data only for the IP addresses that generated fraudulent activity, along with log entries for ALL inbound Google clicks.)
   • Be prepared to “pull the plug” on any PPC company that won’t cooperate: If you encounter substantial suspicious activity in your campaigns with a PPC search system, and the company does not cooperate in a meaningful way, you need to quickly terminate your campaigns with that company. How the company reacts will tell you a lot: when I cancelled my campaigns with company “E,” above, the company simply accepted it without question — they knew I had “caught” them and they just moved on to other victims. In contrast, whenever I suspend my campaigns on Google’s AdWords system, the company immediately asks why, and seeks to identify and remove any obstacles to the continuation of our relationship.

---

This article was written by Mark J. Welch, Internet Marketing Consultant

Comments are closed