`
`
`
`
`DART WHITE PAPER
`
`COUNTING
`
`METHODOLOGIES
`
`“\
`\\
`
`\g
`
`7777777777777777777
`
`,,,,,,,,,,,,,,,,,,,,
`
`JULY 12, 2001
`
`I 5—
`
`J
`
`Click
`
`COPYRIGHT © 2001. DOUBLECLICK INC.
`
`AHBLT-2008.001
`
`AHBLT-2008.001
`
`

`

`TABLE OF CONTENTS
`
`Preface to DART Counting Methodologies.............................ii
`Counting Impressions ..............................................................1
`The Ad Serving Process.........................................................2
`Why Impressions May Be Counted Differently ....................3
`Counting Clicks.........................................................................6
`Process of How DART Counts Clicks ...................................7
`Invalid Clicks .........................................................................11
`Why Clicks May Be Counted Differently.............................14
`Using Referrers to Count Clicks..........................................16
`Counting Unique Users ..........................................................18
`Industry Standards .................................................................20
`Appendix I: DoubleClick Privacy Policy ...............................21
`Appendix II: Counting and WebTrends Log Analyzer .........24
`Discrepancies Between WebTrends Log Analyzer and
`DART ..................................................................................25
`Using Companion Hits to Minimize Discrepancies ...........28
`Why Discrepancies Can Still Occur ....................................30
`Index.........................................................................................31
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`i
`
`AHBLT-2008.002
`
`

`

`About this white
`paper
`
`How this white
`paper is organized
`
`PREFACE TO DART COUNTING
`METHODOLOGIES
`
`Because of a lack of standards within the Internet advertising industry, there is little
`uniformity in the way websites and ad-serving technologies count impressions and
`clicks. As a result, advertisers have difficulty making accurate comparisons among
`the many websites with which they do business.
`
`While third-party ad serving through DoubleClick presents advertisers with an
`accurate, uniform methodology for counting and reporting impressions and clicks
`across multiple websites, there are often discrepancies between the numbers reported
`by DoubleClick and those reported by the internal or third-party counting systems
`used by the individual websites. Furthermore, because there are many factors that can
`cause these discrepancies, and because these factors vary by website, the
`discrepancies are typically not consistent from website to website.
`
`The purpose of this white paper is to explain the discrepancies between the way
`DART and other organizations count impressions and clicks, and to provide ways to
`minimize the impact of these discrepancies.
`
`This white paper is divided into the following sections:
`• Counting Impressions on page 1 explains the ad serving process and discusses
`why websites might count impressions differently from DART.
`• Counting Clicks on page 6 explains the process by which DART counts clicks,
`discusses invalid clicks, explains why websites might count clicks differently from
`DART, and discusses the issue of using referrers to count clicks.
`• Counting Unique Users on page 18 explains how DART counts unique users who
`visit DART-enabled websites.
`Industry Standards on page 20 explains DoubleClick’s approach to industry
`standards for counting impressions and clicks.
`• Appendix I: DoubleClick Privacy Policy on page 21 explains counting
`discrepancies between WebTrends Log Analyzer and DART, and provides
`techniques to minimize those discrepancies.
`• Appendix II: Counting and WebTrends Log Analyzer on page 24 explains
`DoubleClick’s privacy policy and provides sample text for you to use in your
`websites.
`
`•
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`ii
`
`AHBLT-2008.003
`
`

`

`COUNTING IMPRESSIONS
`
`In online advertising, it is imperative to know how many impressions a given ad has
`served. This chapter explains how DART counts impressions as part of the ad serving
`process, and why DART’s impression counts are sometimes different from the
`numbers that are produced by other, non-DoubleClick counting methodologies.
`
`This chapter discusses the following topics:
`
`• The Ad Serving Process on page 2
`• Why Impressions May Be Counted Differently on page 3.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`1
`
`AHBLT-2008.004
`
`

`

`Counting Impressions
`DART Counting Methodologies
`
`
`The Ad Serving Process
`
`The following figure shows the process by which DART serves ads.
`
`
`
`
`
`Web Sewer
`
`
`Firewall
`
`Proxy Server]
`
`FIGURE: How DART serves ads
`
`DoubleClick Data Center
`(Ad and Media Servers)
`
`— The user requests an HTML page.
`The Web server sends the user an HTML document.
`
`The ad server, a media server, or a third-party redirect sewer serves the ad.
`
`3
`
`The user’s browser reads the HTML code and sends a request to keep. //
`nd.doub1oc1ick.n¢t: for an ad.
`
`— The DoubleClick ad server chooses an ad and counts an impression.
`
`DART counts an impression when the ad server determines which ad to send (Stage
`4) after receiving a request for an ad from the user ’5 browser (Stage 3).
`
`Many websites and ad-serving systems count an impression when the HTML page
`that contains the tags for the ad is sent to the user (Stage 2). This counting method
`incorrectly equates a page view with an ad impression. For a variety of reasons,
`browsers oflen send requests for HTML pages, but do not send requests for the ads in
`those pages. When counting impressions, DART counts only requests for ads
`themselves, not requests for the HTML pages that contain the ads.
`
`
`
`DorbleClick DART
`
`Last Modified July 12, 2001
`
`2
`
`AHBLT-2008.005
`
`AHBLT-2008.005
`
`

`

`DART Counting Methodologies
`
`Counting Impressions
`
`Why Impressions May Be Counted Differently
`In order to provide advertisers with the most useful information available,
`DoubleClick has designed its impression-counting methodology to be as accurate as
`possible within current HTML and HTTP restraints.
`
`Websites that use their own counting methodologies may count either more or fewer
`impressions than DART.
`
`This section discusses the following topics:
`
`• Overcounting of impressions by websites on page 3
`• Undercounting of impressions by websites on page 4
`• Techniques to defeat caching on page 4
`• Conclusions about impression discrepancies on page 5.
`
`Overcounting of
`impressions by
`websites
`
`Once a browser has received the HTML for a Web page, it begins reading it from top
`to bottom, requesting each image referenced in the HTML. Among the images the
`browser requests is the ad image.
`
`More often than not, websites and ad-serving systems count impressions as the
`number of page views rather than the number of requests for ads. If the browser
`receives the HTML page (in the Figure on page 2, see Stage 1 on page 2 and Stage 2
`on page 2), but is then prevented from requesting the ad (Stage 3 on page 2), the
`website counts an impression, but DART does not.
`
`There are numerous reasons why browsers do not request all of the images in an
`HTML page:
`
`•
`
`If the user has images turned off in the browser, no images are requested.
`If a DART-enabled website is tagged to accept rich media, the creative is requested
`even if images are turned off in the browser and no image is displayed. Clicks on
`creatives served with rich media tags are likewise counted even if images are
`turned off in the browser and the ad image is not displayed.
`If the DART HTML tags are positioned below tags for other content or images, or
`toward the bottom of an HTML page, it is possible for a user to press Stop or Back
`or click on a link to another page before the ad is requested from DoubleClick.
`• Because of a bug, certain browsers request HTML pages twice, while only
`requesting images once. Websites that count impressions as page views therefore
`incorrectly count two impressions. DART counts only one impression.
`
`•
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`3
`
`AHBLT-2008.006
`
`

`

`DART Counting Methodologies
`
`Counting Impressions
`
`Undercounting of
`impressions by
`websites
`
`Techniques to
`defeat caching
`
`• Sometimes when users are forced to wait for HTML pages because of slow
`Internet service, they repeatedly click Stop and/or Reload or Refresh in order to
`establish a faster connection. In such situations, the website’s server receives
`multiple requests for the HTML page. For each request, the website counts an
`impression, even though users never received an ad. Because the request for an ad
`is not sent to DART until the browser has fully loaded the HTML page and begins
`requesting the images in it, DART does not count any impressions.
`
`Browser and proxy server caching are the primary factors that cause some websites to
`count fewer impressions than DART.
`
`Caching is a technique used to increase performance on the Internet. Web browsers
`and proxy servers save copies of Web page elements (for example, page source code,
`images, and so on) to local memory caches so that they do not have to retrieve the
`same material on subsequent requests. When a browser encounters a URL it has seen
`before, it can retrieve a copy of the Web page directly from the cache, rather than
`sending a new request across the Internet. This decreases the time it takes to deliver
`the requested information to the end user, and reduces the load on Web servers and
`the bandwidth they use.
`
`Caching affects both publishers and advertisers. When Web pages and advertisements
`are requested and retrieved from cache, publishers and ad serving companies have
`difficulty monitoring these requests. Many systems count an impression and a page
`view when a Web page is originally requested by a proxy server (in the Figure on
`page 2 see Stage 1 on page 2 and Stage 2 on page 2), but do not count subsequent
`impressions or page views by users who are accessing the page from the proxy server
`cache. Thus the number of impressions for an ad is often undercounted, as are page
`views.
`
`DoubleClick has developed a tagging and delivery system that effectively defeats
`browser and proxy server caching of ads. Instead of only counting the initial page
`view, DART counts an impression each time the Web page is viewed, even if it is
`served from a browser or proxy cache. This methodology leads to higher, more
`accurate impression counts.
`
`DART uses three techniques to defeat browser and proxy server caching.
`Unique numbers: Random, or unique, numbers are appended to the end of each
`DART HTML tag. Because the unique number makes each image URL unique,
`browsers and proxy servers do not recognize the URLs as ones they have seen before.
`Instead, browsers and proxy servers must request a new ad image each time the page
`is loaded, even if the ad image is already stored in the browser or proxy server cache.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`4
`
`AHBLT-2008.007
`
`

`

`DART Counting Methodologies
`
`Counting Impressions
`
`HTTP 302 commands: DART uses an HMTL protocol known as an HTTP 302
`command or location redirect, which forces browsers and proxy servers to send a
`request to the DART ad servers for each user and each page view, allowing DART to
`record the impressions accurately each time the Web page is viewed. Browsers and
`proxy servers cannot cache HTTP 302 commands, but instead must send a new
`request each time they receive a HTTP 302 command.
`
`Question marks: A question mark (?) is appended to the end of each ad request
`URL in the DART HTML tags. Many proxy servers recognize tags with question
`marks as elements that must not be cached. Appending question marks thus further
`reduces the effects of proxy server caching on the counting of impressions.
`
`If a website does not use similar techniques to defeat caching, HTML pages in the
`website can be stored in and served from browser or proxy server caches. In such
`cases, the website does not record an impression, because no request is sent to the
`website (Step 1 on page 2 and Step 2 on page 2 in the Figure on page 2 above are
`skipped). However, because a new request is sent to the DART ad servers (Step 3 on
`page 2) even when the HTML page is loaded from the cache, DART counts an
`impression (Step 4 on page 2).
`
`For more information on browser and proxy caching, see the Webmaster User Guide.
`
`Because the factors that affect undercounting and overcounting of impressions vary
`from website to website, some websites, such as those that are subject to heavy proxy
`server caching, may report fewer impressions than DART, while others, such as those
`that place ads at the bottom of large or content-heavy Web pages, may report more
`impressions than DART.
`
`DoubleClick has developed a methodology that provides the closest approximation
`possible within current HTML and HTTP constraints of the number of times users
`have actually seen an ad.
`
`Websites can minimize discrepancies in the numbers of impressions counted by:
`
`•
`•
`
`counting ad requests instead of page views
`implementing techniques to defeat caching.
`
`Conclusions about
`impression
`discrepancies
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`5
`
`AHBLT-2008.008
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`COUNTING CLICKS
`
`DoubleClick has taken a leadership position in the online advertising industry by
`implementing a click-counting methodology that is designed to count only clicks
`generated by humans, and excludes clicks generated by automated programs.
`
`This methodology has sometimes put DoubleClick at a competitive disadvantage
`because our websites and advertisers report fewer clicks than websites and advertisers
`whose ad-serving systems include clicks generated by automated programs. However,
`because we believe that for both direct-marketing and brand advertisers, a human
`click is the only click of value — after all, an automated program cannot become a
`customer — we that full disclosure and long-term industry standards are more
`beneficial than short-term competitive gain. We believe that as industry standards
`continue to develop, advertisers will increasingly demand the accuracy provided by
`removing automated clicks from click counts, and that websites will increasingly
`comply.
`
`This chapter discusses the following topics:
`
`• Process of How DART Counts Clicks on page 7
`Invalid Clicks on page 11
`•
`• Why Clicks May Be Counted Differently on page 14
`• Using Referrers to Count Clicks on page 16.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`6
`
`AHBLT-2008.009
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`Process of How DART Counts Clicks
`The following figure shows an example of the process that occurs when a user clicks
`on an ad.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`7
`
`AHBLT-2008.010
`
`

`

`DART Counting Methodologies
`Counting Clicks
`
`
`
`
`
`Publisher's
`Web Server
`
`I
`
`“ ’
`
`.
`
`r:
`
`I
`
`.
`-«
`
`1. User goes to
`website htcp - / /
`.
`'
`www. Slte . com
`
`2. Site receives
`request, sends
`Web page With
`ads
`
`10Adserver
`receives get
`request
`
`3. User clicks on
`ad
`
`Advertiser's
`Web Server
`
`16: Advertiser's
`Site receives
`HTTPget
`request
`
`lghggrzg'fis:
`.1ndex. html
`
`11. Ad server
`5. Site receives
`request
`counts the click
`
`9
`
`4‘ Browser sends
`
`
`6, Site courts the
`HTTPget
`
`click
`request to click-
`
`through URL
`
`8‘ Browser
`receives HTTP
`302 command.
`
`7‘ Site redirects
`browser to
`DoublcClick
`
`9‘ Browser sends
`an HTTP get
`requestto
`DoublcClick
`
`13, Browser
`receives H‘I'I'P
`302 commando
`
`14‘ Browser sends
`an H'ITP get
`request to site
`
`17. User receives
`the file
`index . html,
`
`FIGURE: How DART counts clicks
`
`
`
`DowleCliok DART
`
`Last Modified July 12, 2001
`
`8
`
`AHBLT-2008.011
`
`
`
`
`
`1.2' Ad server
`redirects browser
`to advertiser's site
`
`
`
`
`
`
`
`AHBLT-2008.011
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`Note:
`
`The figure above assumes that the website counts clicks independently of
`DART. For websites that do not count clicks, Stage 4 through Stage 8, as
`represented in the figure above, do not take place.
`
`When the user in the figure above clicks on the ad:
`
`Stage
`
`Description
`
`Initiating the Process
`
`1
`
`2
`
`3
`
`The user goes to the website http://www.site.com.
`
`The website receives the request and sends the Web page, along with the ads.
`Note: This process is actually several steps, including calls to the DoubleClick ad
`servers; however, these steps are not relevant to the process of counting
`clicks, so they are not included here.
`
`The user clicks on an ad.
`
`How the Website Counts the Click
`Note: For websites that do not track clicks independently of DART, skip to Stage 9 on page 9.
`
`4
`
`5
`
`6
`
`7
`
`8
`
`The browser sends an HTTP get request to the click-through URL specified in
`the HTML tags for the ad slot.
`The click-through URL, http://www.site.com/click.cgi?ad1234, points to
`the server used by the website to count clicks.
`Note: These tags are not standard DART HTML tags. Rather, they are tags used by
`the website to track clicks independently of DART.
`
`The website receives the HTTP get request.
`
`The website counts the click.
`
`The website’s server sends an HTTP 302 command that redirects the browser
`to http://ad.doubleclick.net/jump/site.com.
`
`The user’s browser receives the HTTP 302 command.
`
`How DART Counts the Click
`
`9
`
`10
`
`11
`
`The browser sends an HTTP get request to http://ad.doubleclick.net/
`jump/site.com.
`
`DART receives the HTTP get request.
`
`DART counts the click.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`9
`
`AHBLT-2008.012
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`Stage
`
`Description
`
`Completing the Process
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`The DART ad server sends a 302 command that redirects the user’s browser
`to the destination URL, http://www.advertiser.com/index.html.
`
`The user’s browser receives the HTTP 302 command.
`
`The browser sends an HTTP get request for the destination page, http://
`www.advertiser.com/index.html, to the to the advertiser’s Web server.
`
`The advertiser’s Web server receives the request.
`If www.advertiser.com counts hits (requests for files), a hit is recorded.
`Sometimes there is a discrepancy between the number of clicks DART records and
`the number of hits the destination website records. These discrepancies arise
`because users sometimes click Stop or Back, link to another page, or otherwise
`stop the process at this stage, after DART has counted the click but before the
`destination website has counted a hit.
`
`The advertiser’s Web server sends the file index.html to the browser.
`
`The user receives the file index.html.
`
`Effect of redirects
`
`The amount of data that is exchanged in each redirect is very small (usually one or
`two packets). The entire process takes place in milliseconds and causes no noticeable
`increase in latency.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`10
`
`AHBLT-2008.013
`
`

`

`DART Counting Methodologies
`
`Invalid Clicks
`
`Discarding invalid
`clicks
`
`Identifying unique
`users
`
`Counting Clicks
`
`An invalid click is any click that is not generated by an actual human user.
`
`This section discusses the following topics:
`
`• Discarding invalid clicks on page 11
`• Causes of invalid clicks on page 12
`• Preventing automated clicks on page 12.
`
`DART uses a proprietary algorithm to prevent invalid clicks from falsely inflating
`click and click rate statistics. The algorithm works by discarding, or stripping,
`multiple sequential clicks generated by a single user. Analysis has shown that very
`few clicks by real users are being discarded. DART counts a click each time a user
`clicks on an ad and is sent to a click-through URL. DART considers 10 or more clicks
`by the same user within a four-minute period to be invalid, because such clicks are
`nearly always generated by automated programs, not by actual users.
`
`DART can adjust both the time period measured and the number of clicks counted
`within that time period. These parameters can be modified as frequently as necessary.
`Though currently 10 clicks in four minutes from the same user are considered invalid
`and are stripped, DoubleClick can modify this rate if necessary, as customer usage
`patterns vary. (These parameters can be adjusted for the entire DART system only, not
`for individual ad networks.)
`
`In order to discard multiple sequential clicks, DART must identify each unique user.
`
`Using cookies to identify unique users: To identify unique users, DART sets a
`cookie for each user the first time that user is served an ad. Then, when the user clicks
`on an ad, the DART ad servers use the cookie to identify the user.
`
`The DoubleClick cookie is a small text file written by the DART ad servers to a user’s
`hard drive the first time a user visits a DART-enabled website. The cookie is
`anonymous and is used to store a unique number assigned to a browser. Information
`associated with the cookie is updated on each subsequent visit the browser makes to
`any DART-enabled website. As well as identifying a unique user, data associated with
`the cookie provides the ad servers with information that is used for frequency and
`rotation targeting and audience segment targeting.
`
`For more information about unique users, see Counting Unique Users on page 18 and
`for information about how DART uses cookies, see the DART Targeting
`Methodologies white paper.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`11
`
`AHBLT-2008.014
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`Causes of invalid
`clicks
`
`Using IP addresses to identify unique users: Some users do not accept cookies. In
`such cases, the user’s IP address is used. IP addresses, numbers that identify each
`computer currently in use on the Internet, are assigned in blocks. Because of this,
`DART can differentiate a click to a website generated from anywhere on the Internet,
`including those from within the website’s domain. These self-site visits are not
`considered valid for the purposes of counting clicks (for example, clients can run their
`own spiders as a way to test their websites’ capabilities) and are thus discarded.
`
`There are some extra considerations used in the algorithm to account for large proxy
`servers (for example, America Online). Because proxy servers use the same IP
`address for multiple users, 10 or more clicks can be generated in a four-minute period
`from the same IP address even when the clicks are from different individual users.
`Thus, if several users who do not have DoubleClick cookies click on ads within a
`short period of time, it can appear as if the clicks are invalid. To ensure that these
`valid clicks are not discarded, DART cross-references the user’s IP address with a
`database of Internet service providers. If the service provider is known to use proxy
`servers, the clicks are not discarded.
`
`Analysis reveals that there are three major causes of invalid clicks.
`Spiders and crawlers: Spiders and crawlers are programs that automatically
`explore the World Wide Web, searching documents to build topical, statistical, or
`historical indexes of websites. To search a website, a spider or crawler clicks on each
`link in each page, often repeatedly, to access the content.
`Programs that download entire websites for offline use: Programs can be used to
`download everything from multiple Web pages to a full website for use offline. To do
`so, they must click on each link in the website in order to access the content. These
`programs cannot be prevented from accessing a website.
`Link analyzers: Some websites use link analyzers, which are tools used to verify the
`links in the website. To verify a link, the link analyzer must click on it to ensure that it
`is valid.
`
`When a spider, crawler, download program, or link analyzer goes to a website, it
`generates at least one false click per page. The effect of these utilities depends on the
`layout of the website. Consequently, websites with many pages and links are
`significantly affected, while websites with only a few pages and links are not.
`
`Preventing
`automated clicks
`
`Spiders and crawlers and other automated programs are often (but not always)
`designed so that before they send a request to any server, they first check for a file
`called robots.txt, which sends instructions to the automated program and can tell
`the program not to make any requests to that server. DART uses robots.txt to help
`prevent clicks from spiders, crawlers and other automated programs. This file does
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`12
`
`AHBLT-2008.015
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`not, however, prevent download programs and link analyzers from generating
`automated clicks.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`13
`
`AHBLT-2008.016
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`Why Clicks May Be Counted Differently
`When websites that use a counting methodology other than DART find that the
`number of clicks counted are different, the difference can be that either more or fewer
`clicks than counted by DART.
`
`Undercounting of
`clicks by websites
`
`Undercounting of
`clicks by
`advertisers
`
`Overcounting of
`clicks by websites
`
`Conclusions about
`click
`discrepancies
`
`The following topics contain related information:
`
`• Undercounting of clicks by websites on page 14
`• Undercounting of clicks by advertisers on page 14
`• Overcounting of clicks by websites on page 14
`• Conclusions about click discrepancies on page 14.
`
`When a user clicks on an ad, the request is always redirected to DoubleClick;
`therefore, it avoids browser and proxy caching altogether, because browsers and
`proxy servers do not cache redirect requests. As such, there are almost never instances
`in which DART counts a click that the website does not count. This means that
`DoubleClick’s click counts are rarely higher than the click counts reported by
`websites.
`
`Some advertisers use the referrer to count clicks on their ads, which usually leads to
`undercounting of clicks. For more information, see Using Referrers to Count Clicks
`on page 16.
`
`Websites usually report a higher number of clicks than DART because DART
`discards clicks generated by automated programs, while most websites do not. For
`details, see Invalid Clicks on page 11.
`
`While websites almost never count fewer clicks than DART, they often count more
`clicks. The main reason for these discrepancies is that many websites count non-
`human clicks generated by automated programs, while DART does not. In such cases,
`the size of the discrepancy depends on how frequently the website is spidered,
`downloaded for offline use, or otherwise clicked on by automated programs.
`
`Discarding clicks lowers the click rate, but increases the conversion rate — the
`percentage of clicks that lead to sales, downloads, registrations, or similar activities
`— and provides advertisers with more accurate information about the performance of
`their campaigns.
`
`Websites can minimize discrepancies in the numbers of impressions counted by:
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`14
`
`AHBLT-2008.017
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`• using a robots.txt file (see Preventing automated clicks on page 12)
`•
`implementing an algorithm similar to the one used by DART to discard clicks
`generated by automated programs (see Invalid Clicks on page 11).
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`15
`
`AHBLT-2008.018
`
`

`

`DART Counting Methodologies
`
`Counting Clicks
`
`Using Referrers to Count Clicks
`Some advertisers attempt to count clicks using the referrer — the Web address listed
`in the HTTP header as the location from which the user clicked through to the
`advertiser’s website. Using the referrer to count clicks is extremely unreliable for a
`number of reasons.
`
`This section discusses the following topics:
`
`• Counting referrers from DoubleClick domains on page 16
`• Counting referrers from DoubleClick domains and targeted sites on page 17
`• Conclusions about using referrers to count clicks on page 17.
`
`Counting referrers
`from DoubleClick
`domains
`
`The most obvious way to use referrer to count clicks from DoubleClick ads is to count
`referrers from DoubleClick domains. This method, however, tends to undercount
`clicks by a wide margin, because DoubleClick domains are the referrers for only
`certain types of DART HTML tags.
`
`TABLE: Referrers for different types of DART HTML tags
`
`Tag Type
`
`Standard
`
`Layer
`
`Iframe
`
`Frame
`
`JavaScript
`
`Tag Uses the Format:
`
`The referrer is:
`
`jump
`
`adl
`
`adi
`
`adf
`
`adj
`
`The Web page where the ad is displayed.
`
`The Web page where the ad is displayed.
`
`The iframe source (DoubleClick).
`
`The frame source (DoubleClick).
`
`The Web page where the ad is displayed.
`
`Counting referrers from ads served in iframes: All ad slots that use iframes also
`use either layers or JavaScript, because iframes are compatible only with Microsoft
`Internet Explorer, not with Netscape Navigator. Thus even if an advertiser counts all
`referrers from the iframe, users who click on the ad using Netscape Navigator are not
`counted.
`
`Counting referrers from ads served in frames: While it is possible to count the
`number of referrers with some accuracy for an ad that is served only in frames, it is
`generally difficult for an advertiser to make sure that the ad is not served in other
`ways. For example, even if a targeted site uses frames to serve ads, it is possible for
`that site to change its format without notifying the advertiser. Likewise, it is possible
`for that site to establish child sites that inherit the targeted ad and do not use frames to
`serve ads.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`16
`
`AHBLT-2008.019
`
`

`

`DART Counting Methodologies
`
`Counting referrers
`from DoubleClick
`domains and
`targeted sites
`
`Counting Clicks
`
`Some advertisers try to count clicks more accurately by counting referrers not only
`from DoubleClick domains, but also from the domains of the sites to which they have
`targeted ads. This technique gives inaccurate results, and can lead to both
`undercounting and overcounting of clicks.
`Multiple URLs: Many high-traffic sites divide their Web pages among multiple
`URLs. For example, a search engine portal could use not only http://
`www.searchengine.com and http://www.searchengine.net, but also http://
`live.searchengine.com, http://careers.searchengine.com, and so on. As such,
`it is likely that an advertiser might not recognize every referrer URL from the site.
`
`Child sites: If a site has child sites with different URLs, it is possible for the
`advertiser to undercount the number of referrers, because the advertiser has not
`counted the referrers from the child site.
`Search sites: If a site has a search engine that returns the advertiser for some
`searches, it is impossible for the advertiser to determine whether the referrer is from a
`click on an ad or from a click on a link in a search page. For example, if a user looks
`up the advertiser in AltaVista, then clicks through to the advertiser’s website, the
`referrer is the same as when a user clicks on an ad that gives AltaVista as the referrer.
`
`Multiple ads: If an advertiser has multiple ads targeted to the same site, there is no
`way to determine from the referrer which ad (if any) was clicked on. As such,
`advertisers with more than one ad placement running simultaneously on a single site
`are unable to count referrers accurately.
`
`Multiple sites: For any ad targeted across multiple sites (for example, run-of-
`network ads), it is impossible to count referrers accurately because there is no way to
`determine which site is the source of each referrer with a DoubleClick URL.
`
`Conclusions about
`using referrers to
`count clicks
`
`While counting referrers may at first seem like an effective way to count clicks, it is
`not. For numerous reasons, counting referrers, whether from the websites where ads
`are served, from DoubleClick, or in some combination thereof, is an extremely
`inaccurate and problematic way to track clicks.
`
`Because every click on a DoubleClick ad is redirected to the DART ad servers, DART
`is able to count clicks with great accuracy, which is further improved by
`DoubleClick’s algorithms for discarding invalid clicks. Also, many websites first
`redirect clicks on ads to their own click-counting servers. These methods provide far
`more accurate results than are achieved by counting referrers.
`
`DoubleClick DART
`
`Last Modified July 12, 2001
`
`17
`
`AHBLT-2008.020
`
`

`

`DART Counting Methodologies
`
`Counting Unique Users
`
`COUNTING UNIQUE USERS
`
`Every time DART serves an ad, it identifies the user to whom the ad is served as a
`unique user. This process is completely anonymous. It allows DART to report on the
`number of unique users who saw a