Upromise Savings — At What Cost? updated January 25, 2010

Upromise touts opportunities for college savings. When members shop at participating online merchants, dine at participating restaurants, or purchase selected products at retail stores, Upromise collects commissions which fund college savings accounts.

Unfortunately, the Upromise Toolbar also tracks users’ behavior in excruciating detail. In my testing, when a user checked an innocuously-labeled box promising "Personalized Offers," the Upromise Toolbar tracked and transmitted my every page-view, every search, and every click, along with many entries into web forms. Remarkably, these transmissions included full credit card numbers — grabbed out of merchants’ HTTPS (SSL) secure communications, yet transmitted by Upromise in plain text, readable by anyone using a network monitor or other recording system.

Proof of the Specific Transmissions

I began by running a search at Google. The Upromise toolbar transmissions reported the full URL I requested, including my search provider (yellow) and my full search keywords (green).

POST /fast-cgi/ClickServer HTTP/1.0
User-Agent: upromise/3195/3195/UP23806818/0012
Host: dcs.consumerinput.com
Content-Length: 274
Connection: Keep-Alive

md5=ee593c14f70c1b7f8b3341a91c3e3639&ts=1264045792.140&bua=N%2FA&meth=get&eid=300&fid=NULL&bin=24af1f0
&refererHeader=http%3A%2F%2Fwww%2Egoogle%2Ecom%2F&url=http%3A%2F%2Fwww%2Egoogle%2Ecom%2Fsearch%3Fhl%3D
en%26source%3Dhp%26q%3Di%2Bfeel%2Bsick%26aq%3Df%26aql%26aqi%3Dg10%26oq

HTTP/1.1 200 OK
Date: Thu, 21 Jan 2010 03:49:51 GMT
Server: Apache/2.2.3 (Debian) mod_python/3.3.1 Python/2.5.1 mod_ssl/2.2.3 OpenSSL/0.9.8c
Connection: close
Content-Type: text/html; charset=UTF-8

<capture>
<md5>ee593c14f70c1b7f8b3341a91c3e3639</md5>
<version>1.0</version>
</capture>

I clicked a result — a page on Wikipedia. Transmissions included the full URL of my request (blue) as well as the web search provider (yellow) and keywords (green) that had referred me (red) to this site.

POST /fast-cgi/ClickServer HTTP/1.0
User-Agent: upromise/3195/3195/UP23806818/0012
Host: dcs.consumerinput.com
Content-Length: 304
Connection: Keep-Alive

md5=ee7e3174db149d0d97f51b10db9ac58d&ts=1264045931.921&bua=N%2FA&meth=get&eid=300&fid=NULL&bin=24af1f0
&refererHeader=http%3A%2F%2Fwww%2Egoogle%2Ecom%2Fsearch%3Fhl%3Den%26source%3Dhp%26q%3Di%2Bfeel%2Bsick%
26aq%3Df%26aql%3D%26aqi%3Dg10%26oq%3D&url=http%3A%2F%2Fen%2Ewikipedia%2Eorg%2Fwiki%2FI%5FFeel%5FSick

HTTP/1.1 200 OK
Date: Thu, 21 Jan 2010 03:52:11 GMT
Server: Apache/2.2.3 (Debian) mod_python/3.3.1 Python/2.5.1 mod_ssl/2.2.3 OpenSSL/0.9.8c
Connection: close
Content-Type: text/html; charset=UTF-8

<capture>
<md5>ee7e3174db149d0d97f51b10db9ac58d</md5>
<version>1.0</version>
</capture>

I browsed onwards to Buy.com (grey), where I added an item to my shopping cart and proceeded to checkout. When prompted, I entered a (made-up) credit card number. Buy.com appropriately secured the card number with HTTPS encryption. But, remarkably, Upromise extracted and transmitted the full sixteen-digit card number (yellow) — as well as my (also fictitious) CVV code (green), and expiration date (blue).

POST /fast-cgi/ClickServer HTTP/1.0
User-Agent: upromise/3195/3195/UP23806818/0012
Host: dcs.consumerinput.com
Content-Length: 1936
Connection: Keep-Alive

md5=352732ac9ee0d9e970f3c65d62ed03b1&ts=1264046115.702&bua=N%2FA&meth=post&eid=300&fid=NULL&bin=24af1f0
&refererHeader=https%3A%2F%2Fssl%2Ebuy%2Ecom%2FCO%2FCheckout%2FpaymentOptions%2Easpx&url=https%3A%2F%2F
ssl%2Ebuy%2Ecom%2FCO%2FCheckout%2FpaymentOptions%2Easpx%3F%5F%5FEVENTTARGET%26%5F%5FEVENTARGUMENT%26%5F
%5FVIEWSTATE%3D%2FwEPDwUKMTQ1MjY3MzM2Mw9kFgQCAw9kFgQCAQ8PFgIeCEltYWdlVXJsBV1odHRwczovL2EyNDguZS5ha2FtYW
kubmV0L2YvMjQ4Lzg0NS8xMGgvaW1hZ2VzLmJ1eS5jb20vYnV5X2Fzc2V0cy92Ni9oZWFkZXIvMjAwNi9idXlfbG9nby5naWZkZAICD
w8WAh8ABXdodHRwczovL2EyNDguZS5ha2FtYWkubmV0L2YvMjQ4Lzg0NS8xMGgvaW1hZ2VzLmJ1eS5jb20vYnV5X2Fzc2V0cy92Ni9j
b3JwL2NoZWNrb3V0X3Byb2Nlc3MvY2hlY2tvdXRfM19wYXltZW50X2dyZWVuLmdpZmRkAgUPZBYQAgUPZBYCZg9kFgRmDw8WBB8ABVN
odHRwczovL2EyNDguZS5ha2FtYWkubmV0L2YvMjQ4Lzg0NS8xMGgvaW1hZ2VzLmJ1eS5jb20vYnV5X2Fzc2V0cy9jcy9pbWFnZXMvYm
1sLmdpZh4HVmlzaWJsZWdkZAIBDw8WAh8ABWNodHRwczovL2EyNDguZS5ha2FtYWkubmV0L2YvMjQ4Lzg0NS8xMGgvaW1hZ2VzLmJ1e
S5jb20vYnV5X2Fzc2V0cy9idXR0b25zLzIwMDgvYnV0dG9uX2NvbnRpbnVlMi5naWZkZAIHD2QWAmYPZBYKAgUPEA9kFgIeB29uY2xp
Y2sFKFNldFVuaXF1ZVJhZGlvQnV0dG9uKCdyZG9QYXltZW50cycsdGhpcylkZGQCBw8QD2QWAh4Ib25DaGFuZ2UFG2phdmFzY3JpcHQ
6c2hvd0hpZGVDQyh0aGlzKWRkZAIJDw9kFgIfAgUfc3dpdGNoUmFkaW8oJ3Jkb05ld0NyZWRpdENhcmQnKWQCDQ8QZA8WDWYCAQICAg
MCBAIFAgYCBwIIAgkCCgILAgwWDRAFBDIwMTAFBDIwMTBnEAUEMjAxMQUEMjAxMWcQBQQyMDEyBQQyMDEyZxAFBD%2A%2A%2A%2A%7B
1422%7D%26%5F%5FEVENTVALIDATION%3D%2FwEWKQLNvonpCgKYm5r9BALB5eOPBALy2ZqvCQLv2ZqvCQLq2ZqvCQLu2ZqvCQLr2ba
vCQL28Nu2CwLDqZ7xDALDqZrxDALDqabxDALDqaLxDALDqa7xDALDqarxDALDqbbxDALDqfLyDALDqf7yDALcqZLxDALcqZ7xDALcqZ
rxDALrtZj9AgLrtaSgCQLrtbAHAuu13OoIAuu16NEPAuu19LQGAuu1gJgNAuu1rP8FAuu1%2BJcHAuu1hPsPAvai%2BtMMAvaihrcDA
vaikpoKAt7CxckKAsqi76gMAva5sdkEAvLqvYoEAoaI4c0FArr3pYIHApibrtgNijCSRDzGArpVaF4m3bbOLp8yvUM%3D%26rdoPaym
ents%3DrdoNewCreditCard%26lstCCType%3D9%26txtNewCCNumber%3D4412124112341234%26lstNewCCMonth%3D01%26lstN
ewCCYear%3D2010%26txtNewCVV2%3D222%26btnContinue2%2Ex%3D18%26btnContinue2%2Ey%3D19

HTTP/1.1 200 OK
Date: Thu, 21 Jan 2010 03:55:15 GMT
Server: Apache/2.2.3 (Debian) mod_python/3.3.1 Python/2.5.1 mod_ssl/2.2.3 OpenSSL/0.9.8c
Connection: close
Content-Type: text/html; charset=UTF-8

<capture>
<md5>352732ac9ee0d9e970f3c65d62ed03b1</md5>
<version>1.0</version>
</capture>

Upromise also transmitted my email address. For example, when I logged into Restaurant.com, Upromise’s transmission included my email (yellow):

POST /fast-cgi/ClickServer HTTP/1.0
User-Agent: upromise/3195/3195/UP23806818/0012
Host: dcs.consumerinput.com
Content-Length: 378
Connection: Keep-Alive

md5=26830cfab132bb3122fcf40d8cd0f2f9&ts=1264049936.327&bua=N%2FA&meth=post&eid=303&fid=NULL&bin=24af1f0
&refererHeader=&url=https%3A%2F%2Fwww%2Erestaurant%2Ecom%2Fregister%2Dlogin%2Easp%3Fpurchasestatus%3DRE
GISTER%2DLOGIN%26hdnButton%3DSignin%26txtMemberSignin%3Dedelman%40pobox%2Ecom%26radio%5Fcust%3Dcustomer
%5Fexisting%26txtMemberPassword…

HTTP/1.1 200 OK
Date: Thu, 21 Jan 2010 04:58:56 GMT
Server: Apache/2.2.3 (Debian) mod_python/3.3.1 Python/2.5.1 mod_ssl/2.2.3 OpenSSL/0.9.8c
Connection: close
Content-Type: text/html; charset=UTF-8

<capture>
<md5>26830cfab132bb3122fcf40d8cd0f2f9</md5>
<version>1.0</version>
</capture>

All the preceding transmission were made over my ordinary Internet connection just as shown above. In particular, these transmissions were sent in plain text — without encryption or encoding of any kind. Any computer with a network monitor (including, for users connected by Wi-Fi, any nearby wireless user) could easily read these communications. With no additional hardware or software, a nearby listener could thereby obtain, e.g., users’ full credit card numbers — even though merchants used HTTPS security to attempt to keep those numbers confidential.

The Destination of Upromise’s Transmissions: Compete, Inc.

As shown in the "host:" header of each of the preceding communications, transmissions flow to the consumerinput.com domain. Whois reports that this domain is registered to Boston, MA traffic-monitoring service Compete, Inc. Compete’s site promises clients access to "detailed behavioral data," and Compete says more than 2 million U.S. Internet users "have given [Compete] permission to analyze the web pages they visit."

Upromise’s Disclosures Misrepresent Data Collection and Fail to Obtain Consumer Consent

Upromise’s installation sequence does not obtain users’ permission for this detailed and intrusive tracking. Quite the contrary: Numerous Upromise screens discuss privacy, and they all fail to mention the detailed information Upromise actually transmits.

The Upromise toolbar installation page touts the toolbar’s purported benefits at length, but mentions no privacy implications whatsoever.

If a user clicks the prominent button to begin the toolbar installation, the next screen presents a 1,354-word license agreement that fills 22 on-screen pages and offers no mechanism to enlarge, maximize, print, save, or search the lengthy text. But even if a user did read the license, the user would receive no notice of detailed tracking. Meanwhile, the lower on-screen box describes a "Personalized Offers" feature, which is labeled as causing "information about [a user’s] online activity [to be] collected and used to provide college savings opportunities" But that screen nowhere admits collecting users’ email addresses or credit card numbers. Nor would a user rightly expect that "information about … online activity" means a full log of every search and every page-view across the entire web.

The install sequence does link to Upromise’s privacy policy. But this page also fails to admit the detailed tracking Upromise performs. Indeed, the privacy policy promises that Personalized Offers data collection will be "anonymous" — when in fact the transmissions include email addresses and credit card numbers. The privacy policy then claims that any collection of personal information is "inadvertent" and that such information is collected only "potentially." But I found that the information transmissions described above were standard and ongoing.

The privacy policy also limits Upromise’s sharing of information with third parties, claiming that such sharing will include only "non-personally identifiable data." But I found that Upromise was sharing highly sensitive personal information, including email addresses and credit card numbers.

In addition, Upromise’s data transmissions contradict representation in Upromise’s FAQ. The top entry in the FAQ promises that Upromise "has implemented security systems designed to keep [users’] information safe." But transmitting credit card numbers in cleartext is reckless and ill-advised — specifically contrary to standard Payment Card Industry (PCI) rules, and the opposite of "safe." Indeed, in an ironic twist, Upromise’s FAQ goes on to discuss at great length the benefits of SSL encryption — benefits of course lacking when Upromise transmits users’ credit card numbers without such encryption.

The Upromise toolbar offers an Options screen which again makes no mention of key data transmissions. The screen admits collecting "information about the web sites you visit." But it says nothing of collecting email addresses, credit card numbers, and more.

The Scope and Solution

The transmissions at issue affect only those users who agree to run Upromise’s Personalized Offers system. But until last week, that option was set as the default upon new Upromise toolbar installations — inviting users to initiate these transmissions without a second look.

Even if Upromise ceased the most outrageous transmissions detailed above, Upromise’s installation disclosures would still give ample basis for concern. To describe tracking, transmitting, and analyzing a user’s every page-view and every search, Upromise’s install screen euphemistically mentions that its "service provider may use non-personally identifiable information about your online activity." This admission appears below a lengthy EULA, under a heading irrelevantly labeled "Personalized Offers" — doing little to draw users’ attention to the serious implications of Personalized Offers. I don’t see why Upromise users would ever want to accept this detailed tracking: It’s entirely unrelated to the college savings that draws users to the Upromise site. But if Upromise is to keep this function, users deserve a clear, precise, and well-labeled statement of exactly what they’re getting into.

Upromise’s multi-faceted business raises other concerns that also deserve a critical look. The implications for affiliate merchants are particularly serious — a risk of paying affiliate commission for users already at a merchant’s site. But I’ll save this subject for another day.

Upromise’s Response (posted: January 23, 2010)

Upromise told PC Magazine’s Larry Seltzer that less than 2% of their 12 million members were affected. Though 2% of 12 million is 240,000 — a lot!

Upromise staff also wrote to me. I replied with a few questions:

1) How long was the problem in effect?

2) How many users were affected?

3) What caused the problem?

4) I notice that the Upromise toolbar EULA has numerous firmly-worded defensive provisions. (See the entire sections captioned “Disclaimer of Warranty” and “Limitation of Liability” – purporting to disclaim responsibility for a wide variety of problems. And then see Governing Law and Arbitration for a purported limitation on what law governs and how claims may be brought.) For consumers who believe they suffered losses or harms as a result of this problem, will Upromise invoke the defensive provisions in its EULA to attempt to avoid or limit liability?

I’m particularly interested in the answer to the final question. The EULA is awfully one-sided. I appreciate Upromise confirming the serious failure I uncovered. Upromise should also accept whatever legal liability goes with this breach.

Upromise’s Response to My Questions (posted: January 25, 2010)

Upromise replied to my four questions to indicate that the scope of the problem was "approximately 1% of our 12 million members." Upromise did not answer my questions about the duration of the problem, the cause of the problem, or the effect of Upromise’s EULA defenses.

Google Click Fraud Inflates Conversion Rates and Tricks Advertisers into Overpaying

I’ve repeatedly reported improper placements of Google ads. In most of my write-ups, the impropriety occurs in ad placement — Google PPC ads shown in spyware popups (1, 2, 3, 4), in typosquatting sites (1, 2), or in improperly-installed and/or deceptive toolbars (1, 2). This article is different: Here, the impropriety includes a fake click — click fraud — charging an advertiser for a PPC click, when in fact the user never actually clicked.

But this is no ordinary click fraud. Here, spyware on a user’s PC monitors the user’s browsing to determine the user’s likely purchase intent. Then the spyware fakes a click on a Google PPC ad promoting the exact merchant the user was already visiting. If the user proceeds to make a purchase — reasonably likely for a user already intentionally requesting the merchant’s site — the merchant will naturally credit Google for the sale. Furthermore, a standard ad optimization strategy will lead the merchant to increase its Google PPC bid for this keyword on the reasonable (albeit mistaken) view that Google is successfully finding new customers. But in fact Google and its partners are merely taking credit for customers the merchant had already reached by other methods.

In this piece, I show the details of the spyware that tracks user browsing and fakes Google PPC ad clicks, and I identify the numerous intermediaries that perpetrate these improper charges. I then criticize Google’s decision to continue placing ads through InfoSpace, the traffic broker that connected Google to this click fraud chain. I consider this practice in light of Google’s advice to advertisers and favored arguments that click fraud problems are small and manageable. Finally, I propose specific actions Google should take to satisfy to prevent these scams and to satisfy Google’s obligations to advertisers.

Introducing the Problem: A Reader’s Analogy

Reading a prior article on my site, a Register discussion forum participant offered a useful analogy:

Let’s say a restaurant decides [it] wants someone to hand out fliers … so they offer this guy $0.10 a flier to print some and distribute them.

The guy they hire just stands at the front door and hand the fliers to anyone already walking through the door.

Restaurant pays lots of money and gains zero customers.

Guy handing out the fliers tells the owner how many fliers were printed and compares that to how many people bring the fliers into his restaurant.

The owner thinks the fliers are very successful and now offers $0.20 for each one.

It’s easy to see how the restaurant owner could be tricked. Such scams are especially easy in online advertising — where distance, undisclosed partnerships, and general opacity make it far harder for advertisers to figure out where and how Google and its partners present advertisers’ offers.

Google and Its Partners Covering Advertisers’ Sites with Spyware-Delivered Click-Fraud Popups

PPC advertisers (e.g. Finish Line)
money viewers
   Google   
money viewers
InfoSpace
money viewers
Cheapstuff
money viewers
Adfirmative
money viewers
dSide Marketing
money viewers
Netaxle
money viewers
eWoss
money viewers
AdOn Network
money viewers
Trafficsolar

The money trail – how funds flow from advertisers to Google to Trafficsolar spyware.

In testing of December 31, 2009, my Automatic Spyware Advertising Tester browsed Finishline.com, a popular online shoe store, on a virtual computer infected with Trafficsolar spyware (among other advertising software, all installed through security exploits without user consent). Trafficsolar opened a full-screen unlabeled popup, which ultimately redirected back to Finish Line via a fake Google PPC click (i.e., click fraud).

My AutoTester preserved screenshots, video, and packet log of this occurrence. The full sequence of redirects:

Trafficsolar opens a full-screen popup window loading from urtbk.com, a redirect server for AdOn Network. (AdOn, of Tempe, Arizona, first caught my eye when it boasted of relationships with 180solutions/Zango and Direct Revenue. NYAG documents later revealed that AdOn distributed more than 130,000 copies of Direct Revenue spyware. More recently, I’ve repeatedly reported AdOn facilitating affiliate fraud, inflating sites’ traffic stats, and showing unrequested sexually-explicit images.)

AdOn redirects to eWoss. (eWoss, of Overland Park, Kansas, has appeared in scores of spyware popups recorded by my testing systems.)

eWoss redirects to Netaxle. (NetAxle, of Prairie Village, Kansas, has also appeared in numerous popups — typically, as here, brokering traffic from eWoss.)

Netaxle redirects to dSide Marketing. (dSide Marketing, of Montreal, Canada, says it provides full-service SEO and SEM services.)

dSide Marketing redirects to Adfirmative. (Adfirmative, of Austin, Texas, promises “click-fraud protected, targeted advertising” and “advanced click-fraud prevention.”)

Adfirmative redirects to Cheapstuff. (Cheapstuff fails to provide an address on its web site or in Whois, though its posted phone number is in Santa Monica, California. Cheapstuff’s web site shows a variety of commercial offers with a large number of advertisements.)

Cheapstuff redirects to InfoSpace. (InfoSpace, of Bellevue, Washington, is discussed further in the next section.)

InfoSpace redirects to Google, which redirects through DoubleClick and onwards back to Finish Line — the same site my tester had been browsing in the first place.

This placement is a bad deal for Finish Line for at least two reasons. First, Google charges Finish Line a fee to access a user already at Finish Line’s site. But that’s more of a shake-down then genuine advertising: an advertiser should not have to pay to reach a user already at its site. Furthermore, Google styles its advertising as “pay per click”, promising advertisers that “You’re charged only if someone clicks your ad.” But here, the video and packet log clearly confirm that the Google click link was invoked without a user even seeing a Google ad link, not to mention clicking it. Advertisers paying high Google prices deserve high-quality ad placements, not spyware popups and click fraud.

Finally, the popup lacks the labeling specifically required by FTC precedent. Consistent with FTC’s settlement in its Direct Revenue and Zango cases, every spyware/adware popup must be labeled with the name of the program that caused the popup, along with uninstall instructions. Furthermore, the FTC has taken an appropriately dim view of advertising software installed on users’ computers without user consent. But every single Trafficsolar installation I’ve ever seen has arrived on my test computers through security exploits, without consent. For these reasons, this Trafficsolar-Google popup clearly falls afoul of applicable FTC requirements.

Critiquing InfoSpace’s role

As shown in the prior section and diagram, traffic flows through a remarkable seven intermediaries en route from Trafficsolar spyware to the victim Google advertiser. Looking at such a lengthy chain, the problem may seem intractable: How could Google effectively supervise a partner’s partner’s partner’s partner’s partner’s partner’s partner’s partner? That insurmountable challenge is exactly why Google should never have gone down this path. Instead, Google should place ads only through the companies with which Google has direct relationships.

In this instance, when traffic finally gets to Google, it comes through a predictable source: InfoSpace. It was InfoSpace, and InfoSpace alone, that distributed Google ads into the morass of subsyndicators and redistributors detailed above.

Flipping through my records of prior InfoSpace observations, I was struck by the half-decade of bad behavior. Consider:

June 2005: I showed InfoSpace placing Google ads into the IBIS Toolbar which, I demonstrated in multiple screen-capture videos, was arriving on users’ computers through security exploits (without user consent). The packet log revealed that traffic flowed from IBIS directly to InfoSpace’s Go2net.com — suggesting that InfoSpace had a direct relationship with IBIS and paid IBIS directly, not via any intermediary.

August 2005: I showed InfoSpace placing ads through notorious spyware vendor Direct Revenue (covering advertisers’ sites with unlabeled popups presenting their own PPC ads). The packet log revealed that traffic flowed from Direct Revenue directly to InfoSpace — suggesting that InfoSpace had a direct relationship with Direct Revenue and paid Direct Revenue directly, not via any intermediary.

August 2005: I showed InfoSpace placing ads through notorious spyware vendor 180solutions/Zango. The packet log revealed that traffic flowed from 180solutions directly to InfoSpace — suggesting that InfoSpace had a direct relationship with 180solutions and paid 180solutions directly, not via any intermediary.

February 2009: I showed InfoSpace placing Google ads into WhenU popups that covered advertisers’ sites with their own PPC ads.

May 2009: Again, I showed InfoSpace using WhenU to cover advertisers’ sites with their own PPC ads, through partners nearly identical to the February report.

January 2010 (last week): I showed InfoSpace’s still placing Google ads into WhenU popups and still covering advertisers’ sites with their own PPC ads.

And those are just placements I happened to write up on my public site! Combine this pattern of behavior with InfoSpace’s well-documented accounting fraud, and InfoSpace hardly appears a sensible partner for Google and the advertisers who entrust Google to manage their spending.

Nor can InfoSpace defend this placement by claiming Cheapstuff looked like a suitable place to show ads. The Cheapstuff site features no mailing address or indication of the location of corporate headquarters. WHOIS lists a “privacy protection” service in lieu of a street address or genuine email address. These omissions are highly unusual for a legitimate advertising broker. They should have put InfoSpace and Google on notice that Cheapstuff was up to no good.

This Click Fraud Undercuts Google’s Favorite Defense to Click Fraud Complaints

When an advertiser buys a pay-per-click ad and subsequently makes a sale, it’s natural to assume that sale resulted primarily from the PPC vendor’s efforts on the advertiser’s behalf. But the click fraud detailed in this article takes advantage of this assumption by faking clicks to target purchases that would have happened anyway. Then, when advertisers evaluate the PPC traffic they bought, they overvalue this “conversion inflation” traffic — leading advertisers to overbid and overpay.

Indeed, advertisers’ following Google’s own instructions will fall into the overbidding trap. Discussing “traffic quality” (i.e. click fraud and similar schemes),Google tells advertisers to “track campaign performance” for “ROI monitoring.” That is, when an advertiser sees a Google ad click followed by a sale, the advertiser is supposed to conclude that ads are working well and delivering value, and that click fraud is not a problem. Google’s detailed “Click Fraud: Anecdotes from the Front Line” features a similar approach, advising that “ROI is king,” again assuming that clicks that precede purchases must be valuable clicks.

Google’s advice reflects an overly optimistic view of click fraud. Google assumes click fraudsters will send random, untargeted traffic. But click-frauders can monitoring user activities to identify the user’s likely future purchases, just as Trafficsolar does in this example. Such a fraudster can fake the right PPC clicks to get credit for traffic that appears to be legitimate and valuable — even though in fact the traffic is just as worthless as other click fraud.

What Google Should Do

Google’s best first step remains as in my posting last week: Fire InfoSpace. Google doesn’t need InfoSpace: high-quality partners know to approach Google directly, and Google does not need InfoSpace to add further subpartners of its own.

Google also needs to pay restitution to affected advertisers. Every time Google charges an advertiser for a click that comes from InfoSpace, Google relies on InfoSpace’s promise that the click was legitimate, genuine, and lawfully obtained. But there is ample reason to doubt these promises. Google should refund advertisers for corresponding charges — for all InfoSpace traffic if Google cannot reliably determine which InfoSpace traffic is legitimate. These refunds should apply immediately and across-the-board — not just to advertisers who know how to complain or who manage to assemble exceptional documentation of the infraction.

More generally, Google must live up to the responsibility of spending other people’s money. Through its Search Network, Google takes control of advertisers’ budgets and decides, unilaterally, where to place advertisers’ ads. (Indeed, for Search Network purchases, Google to this day fails to tell advertisers what sites show their ads. Nor does Google allow opt-outs on a site-by-site basis — policies that also ought to change.) Spending others’ money, wisely and responsibly, is a weighty undertaking. Google should approach this task with significantly greater diligence and care than current partnerships indicate. Amending its AdWords Terms and Conditions is a necessary step in this process: Not only should Google do better, but contracts should confirm Google’s obligation to offer refunds when Google falls short.

I’m disappointed by Google’s repeated refusal to take the necessary precautions to prevent these scams. InfoSpace’s shortcomings are well-known, longstanding, and abundantly documented. What will it take get Google to eject InfoSpace and protect its advertisers’ budgets?

Google Still Charging Advertisers for Conversion-Inflation Traffic from WhenU Spyware updated January 7, 2010

When an advertiser buys a pay-per-click ad and subsequently makes a sale, it’s natural to assume that sale resulted primarily from the PPC vendor’s efforts on the advertiser’s behalf. But tricky PPC platforms take advantage of this assumption by referring purchases that would have happened anyway. Then, when advertisers evaluate the PPC traffic they bought, they overvalue this “conversion inflation” traffic — leading advertisers to overbid and overpay.

In this piece, I show Google and its partners still covering popular sites with PPC advertisements promoting those same sites. I present the role of InfoSpace, the Google partner at the core of these misplacements, and I argue that Google should long ago have severed its ties to InfoSpace. I cite specific Google promises that these placements violate, and I critique Google’s contractual disclaimers that claim advertisers must pay for these bogus placements. Finally, I propose specific actions Google should take to satisfy to its obligations to advertisers.

Google and Its Partners Still Covering Advertisers’ Sites with Spyware-Delivered Popups

WhenU covers Continental with its own Google ads -- charging ad fees for traffic Continental would otherwise receive for free
WhenU covers Continental with its own Google ads — chargingad fees for traffic Continental would otherwise receive for free

As shown in the thumbnail at right and detailed in screenshots, video, and packet log, WhenU continues to cover web sites with PPC popups. Crucially, those popups show Google ads — often promoting the very same sites users are already browsing.

In the example shown at right, I browsed the Continental Airlines site. WhenU opened the popup shown at right — covering the Continental site with a list of Google ads, putting a prominent Continental ad front-and-center. Thus, Google charges Continental a fee to access a user already at Continental’s site. That’s a rotten deal for Continental: For one, an advertiser should not have to pay to reach a user already at its site. Furthermore, advertisers paying high Google prices deserve high-quality ad placements, not spyware popups.

The details of the Continental ad, as shown in the WhenU-Google popup, further entice users to click. The ad promises a “low fare guarantee” — suggesting that users who book some other way (without clicking the ad) may not enjoy that guarantee. And the ad promises to take users to the “official site” — suggesting that users who don’t click the ad will book through a site that is less than official. In fact both suggestions are inaccurate, but a reasonable user would naturally reach these conclusions based on the wording of the advertisement and the context of its appearance.

The WhenU-Google popup lacks the labeling specifically required by FTC policy. In particular, all sponsored search ads are to be labeled as such, pursuant to the FTC ‘s 2002 instructions. But look closely at the popup screenshot. On my ordinary 800×600 screen, no such label appears. I gather the required label would ordinarily appear on a sufficiently large screen, but the FTC’s policies make no exceptions for users with small to midsized screens. Indeed, as netbooks gain popularity, small screens are increasingly common.

The diagram below (left) confirms the specific intermediaries passing traffic from WhenU to Google in this instance.

The money trail: how funds flow from advertisers to Google to WhenU
(three examples persisting over ten months)
December 2009

PPC advertisers
(e.g. Continental)
money viewers
   Google   
money viewers
InfoSpace
money viewers
LocalPages
money viewers
(unknown company*)
money viewers
WhenU

PPC advertisers
(e.g. RCN)
money viewers
   Google   
money viewers
InfoSpace
money viewers

*  LocalPages
money viewers
Nbcsearch
money viewers
LocalPages

money viewers
WhenU

PPC advertisers
(e.g. Verizon)
money viewers
   Google   
money viewers
InfoSpace
money viewers
LocalPages
money viewers
WhenU

This observation marks the third sequence by which I have observed Google paying WhenU to cover advertisers’ sites with the advertisers’ own Google ads. The center and right diagrams (above) show the intermediaries in my May 2009 and February 2009 observations of similar placements.

The Impropriety of Google’s Relationship with InfoSpace

In all three instances I reported (as summarized in the diagram above), Google’s closest link is to InfoSpace. That is, Google pays InfoSpace, and InfoSpace pays the various entities that follow. In my view, Google’s relationship with InfoSpace is ill-advised for at least three reasons:

First, InfoSpace has a track record of improper placements of Google ads. InfoSpace is implicated in all three of the placements detailed above — misplacements that have continued over a lengthy period despite ample notice and opportunity for correction. Furthermore, I have personally observed other improper placements by InfoSpace. (Perhaps I’ll post more in a futher piece.) Google need not continue to do business with a distributor with such a poor track record.

Second, Google does not need a distributor whose business model entails farming out ad placements to subdistributors. If InfoSpace’s subdistributors seek to distribute Google ads, and to be paid for doing so, let them apply directly to Google and undergo Google’s ordinary quality control and oversight. Inserting InfoSpace as an additional intermediary serves only to lessen accountability.

Third, InfoSpace’s corporate history undermines any request for lenience or forgiveness. The Seattle Times chronicles InfoSpace’s accounting fraud in a three-part investigative report, “Dot-Con Job“, presenting 12,000+ words of analysis as well as primary source documents and even voicemail recordings. The Seattle Times byline summarizes their findings: “Investors were cashing out millions, and faithful investors were left with pennies.” Hardly a mark of trustworthiness!

These Ads Violate Google’s Promises to Users

These ad placements fall short of Google’s promises to users. By paying spyware vendors to show advertisements, Google both enlarges and prolongs the spyware problem. In particular, Google’s funding supports software that users struggle to remove from their computers. Google’s payments make it more profitable for vendors to sneak such software onto users’ computers in the first place.

Furthermore, Google’s Software Principles specifically disallow WhenU’s practices. Google’s “installation” and “upfront disclosure” principles disallow deceptive and nonconsensual WhenU installations. (I have video proof on file showing nonconsensual WhenU installations.) Google’s prohibition on “snooping” prohibits certain WhenU privacy practices, including WhenU’s historic violation of its own privacy policy (transmitting full page URLs despite a privacy policy promising “As the user surfs the Internet, URLS visited by the user … are NOT transmitted to WhenU.com or any third party server”).

Crucially, Google’s partnership with WhenU directly contradicts Google’s call for software makers and advertising intermediaries to “keep[] good company” by supervising partners. Despite that commitment, present on Google’s site for 4+ years, Google inexplicably continues its relationship with WhenU.

These Ads Violate Google’s Promises to Advertisers

These ad placements also fall short of Google’s obligations to advertisers. For example, when Google describes its Search Network, Google promises:

Ads are targeted based on a user’s search terms.   (emphasis added)

But here, the user performed no search — so there was no proper cause to display a Search Network ad or charge an advertiser a high Search Network price.

Google confirms:

On the Search Network, ads are shown … on … the search results pages of … Google’s search partners … within the Search Network. On our search partners, your ads may appear alongside or above search results, as part of a results page as a user navigates through a site’s directory, or on other relevant search pages.   (emphasis added)

A placement through a spyware popup does not meet these criteria: A spyware popup is not a “page.” Furthermore, a user browsing an ordinary web site (like the Continental site shown above) is neither “search[ing]” nor navigating a “directory,” contrary to Google’s promise that search ads are shown to users at search engines and directories.

Despite these clear promises, Google’s AdWords Terms and Conditions purport to allow these placements and any others Google might choose to foist on unsuspecting advertisers. Google requires advertisers to accept the following form contract provisions:

Customer understands and agrees that ads may be placed on … (z) any other content or property provided by a third party (‘Partner’) upon which Google places ads (‘Partner Property’).   (emphasis added)

That’s circular, uninformative, and a rotten deal. Advertisers should demand better. Nor should Google’s fine print claim the right to impose such bogus charges. Google should amend its contract to disavow charges from spyware, adware, conversion-inflation, and other schemes contrary to Google’s affirmative promises.

What Google Should Do

Google’s first step is easy: Fire InfoSpace. Google doesn’t need InfoSpace, and there’s zero reason for this relationship to continue in light of InfoSpace’s repeated failings.

Google also needs to pay restitution to affected advertisers. Every time Google charges an advertiser for a click that comes from InfoSpace, Google relies on InfoSpace’s promise that the click was legitimate, genuine, and lawfully obtained. But there is ample reason to doubt these promises. Google should refund advertisers for corresponding charges — for all InfoSpace traffic if Google cannot reliably determine which InfoSpace traffic is legitimate. These refunds should apply immediately and across-the-board — not just to advertisers who know how to complain or who manage to assemble exceptional documentation of the infraction. (Indeed, in response to my May 2009 report, I know Google provided a credit to RCN — the specific advertiser whose targeting I happened to feature in my example. But I gather Google failed to provide automatic credits to all affected advertisers, even though Google’s billing records provide ample documentation of which advertisers faced charges from which Google partners. And I understand that Google denied requests for refunds or credits from other affected advertisers.)

More generally, Google must live up to the responsibility of spending other people’s money. Through its Search Network offering, Google takes control of advertisers’ budgets and decides, unilaterally, where to place advertisers’ ads. (Indeed, for Search Network purchases, Google to this day fails to tell advertisers what sites show their ads. Nor does Google allow opt-outs on a site-by-site basis — policies that also ought to change.) Spending others’ money, wisely and responsibly, is a weighty undertaking. Google should approach this task with significantly greater diligence and care than current partnerships indicate. Amending its AdWords T&C’s is a necessary step in this process: Not only should Google do better, but contracts should confirm Google’s obligation to offer refunds when Google falls short.

I’m disappointed by how little has changed since my year-ago reports of these same practices. In a conference presentation in February 2009, I demonstrated substantially similar WhenU placements, with Google’s Rose Hagan (Senior Trademark Counsel) present in the audience. In May 2009 I wrote up these WhenU placements on my web site in great detail. Yet ten months later, the problem continues unabated. Indeed, the other misplacements I identified in May 2009 also continue: Google continues partnering with IAC SmileyCentral (deceptive browser plug-ins that induce searches when users attempt navigations), placing ads on typosquatting sites (including sites that show a company’s own ads when users mistype that company’s domain name), and, through Google Chrome, inviting users to search (and click prominent top-of-page ads) when direct navigation would better satisfy users’ requests and avoid unnecessary advertising costs for advertisers. I’m disappointed by the lack of progress when, in each instance, the improper charges are clear and well-documented. Google’s intransigence confirms the need for the Bill of Rights for Online Advertisers I proposed this fall.

Measuring the Perpetrators and Funders of Typosquatting

Moore, Tyler, and Benjamin Edelman. “Measuring the Perpetrators and Funders of Typosquatting.” Lecture Notes in Computer Science. Springer-Verlag. Financial Cryptography and Data Security: Proceedings of the International Conference 6052 (2010). (Introduction, Web appendix.)

We describe a method for identifying “typosquatting”, the intentional registration of misspellings of popular website addresses. We estimate that at least 938,000 typosquatting domains target the top 3,264 .com sites, and we crawl more than 285,000 of these domains to analyze their revenue sources. We find that 80% are supported by pay-per-click ads, often advertising the correctly spelled domain and its competitors. Another 20% include static redirection to other sites. We present an automated technique that uncovered 75 otherwise legitimate websites which benefited from direct links from thousands of misspellings of competing websites. Using regression analysis, we find that websites in categories with higher pay-per-click ad prices face more typosquatting registrations, indicating that ad platforms such as Google AdWords exacerbate typosquatting. However, our investigations also confirm the feasibility of significantly reducing typosquatting. We find that typosquatting is highly concentrated: of typo domains showing Google ads, 63% use one of five advertising IDs, and some large name servers host typosquatting domains as much as four times as often as the web as a whole.

Who Owns Metrics?: Building a Bill of Rights for Online Advertisers

Edelman, Benjamin. “Who Owns Metrics?: Building a Bill of Rights for Online Advertisers.” Journal of Advertising Research 49, no. 4 (December 2009). (Adapted from Towards a Bill of Rights for Online Advertisers.)

I offer five rights to protect advertisers from increasingly powerful ad networks-avoiding fraudulent charges for services not rendered, guaranteeing data portability so advertisers get the best possible value, and assuring price transparency so advertisers know what they’re buying. I explain the need for these rights by presenting specific practices causing particular concern.

Red Light States: Who Buys Online Adult Entertainment?

Edelman, Benjamin. “Red Light States: Who Buys Online Adult Entertainment?” Journal of Economic Perspectives 23, no. 1 (Winter 2009): 209-220.

This paper studies the adult online entertainment industry, particularly the consumption side of the market. In particular, it focuses on the demographics and consumption patterns of those who subscribe to adult entertainment websites. On the surface, this business would seem to face a number of obstacles. Regulatory and legal barriers have already been mentioned. In addition, those charging for access to adult entertainment face competition from similar content available without a fee. In the context of adult entertainment, free access offers consumers an extra benefit: online payments tend to create records documenting the fact of a customer’s purchase; consumers of free content may feel more confident that their purchases will remain confidential. More broadly, measured levels of religiosity in America are high. On the other hand, social critics often argue that the rise of Internet pornography is contributing to a coarsening of American culture. Do consumption patterns of online adult entertainment reveal two separate Americas? Or is the consumption of online adult entertainment widespread, regardless of legal barriers, potential for embarrassment, and even religious conviction?

Personal Rapid Transport

This page indexes my writings about Personal Rapid Transport.

Deception in Post-Transaction Marketing updated December 5, 2009

Post-transaction marketers Webloyalty, Vertrue, and Affinion have attracted criticism for solicitations that tend to deceive consumers. They typically feature recurring billing programs that promise a savings or discount, but actually charge users on an ongoing basis. They promote these services while customers are finishing the checkout process at trusted e-commerce sites — a time when few users expect unrelated offers from third parties. Furthermore, they obtain consumers’ credit card numbers from partner sites — so a user may enter a billing relationship and face credit card charges without providing a card number to the company that posts the charges.

In this posting, I present key primary source documents (internal company emails and analyses and reports from victim consumers) as well as outside analyses (a Senate staff report and testimony from hearing witnesses including my own statement for the record).

Higlights of my Statement for the Record: I argue that the timing, placement, and format of post-transaction offers deceptively suggest that the offers are part of the checkout process. (3) I suggest that automatic transfer of consumers’ payment information removes a key warning that customers are incurring a financial obligation. (3-4) I examine disclosures and find them inadequate to cure the deception resulting from the substance, format, and context of the offers. (5) I point out that credit card network rules disallow key post-transaction marketing practices, and I suggest that credit card networks enforce these rules. (6-7) I suggests that low usage rates support an inference of deception, and I provide an empirical strategy to estimate usage rates from publicly-available sources. (7)

Details:

Deception in Post-Transaction Marketing

In a subsequent analysis, I cite, quote, and analyze relevant credit card network rules — finding that those requirements disallow key post-transaction marketing practices:

Payment Card Network Rules Prohibit Aggressive Post-Transaction Tactics

Personal Rapid Transport at Vectus, Ltd. (teaching materials)

Edelman, Benjamin. “Personal Rapid Transport at Vectus, Ltd.” Harvard Business School Case 910-010, November 2009. (Revised September 2010.) (Featured in Working Knowledge: Can Entrepreneurs Drive People Movers to Success?) (educator access at HBP. courtesy copy.)

Personal Rapid Transport (PRT) vehicles—often called “driverless taxis”—sought to combine the best characteristics of cars, taxis, and trains, while adding features unavailable in any existing transportation system. Like cars and taxis, PRT vehicles carried small groups—often just a single passenger—with no need to wait for a shared vehicle to arrive or for others to board. Yet PRT followed train systems in using an exclusive right of way that avoided delays from other traffic. Where would such systems be most useful? Could system designers successfully compete with well-established networks of trains, buses, cars, and roads?

Teaching Materials:

Personal Rapid Transport at Vectus, Ltd. – Teaching Note (HBP 910024)