Search, ads, content.
Judge, jury, executioner?

Some publishers are suspicious about the accuracy of panel-based measurement systems (e.g., comScore, Nielsen NetRatings), believing their count to be grossly underneath the real number of unique visitors to their sites.

Now they have growing reason to suspect Google of worse, even biased, figures.

Panel-based measurement gathers a sample of internet users and records their habits, typically by installing software on their computers that track online activity. The data is then scaled out to project the habits of the entire internet population, which is why panel-measurement companies are careful to use a panel whose diversity accurately reflects that of the country.

In contrast, server-based measurement, tracks about 100% of all traffic (via Javascript tags or by uploaded log files), including "misleading activity" such as internal traffic, duplication, search spiders and robots.

This is why companies like Google often report higher traffic numbers from server logs than panel-based measurement companies do.

Data from its server logs, published through Trends, Ad Planner and Google Analytics, suggests traffic to websites is even lower than what comScore shows — by as much as three times less, writes John Battelle on Searchblog. (See a graph of the data comparison.)

With help from Gian Folgoni at comScore, which juxtaposed Google Ad Planner to comScore across 20,163 websites, Battelle concluded that Google both undercounts unique visitors, and slightly penalizes sites that are not part of the AdSense network (that is, the visitors discrepancy is not so high for them):

This raises the question, "Is Google biased?" An ad planner seeking highly-trafficked sites may be more likely to buy on a site running Google ads. Sites running AdSense will likely also have more accurate traffic figures, since Google naturally tracks more data on those pages.

Not to say server log data is perfect. When consumers delete cookies — the data-trackers installed by websites they visit — they affect traffic figures. Panel metrics are not typically affected by cookie deletion, but server log data is. For example, mlb.com could have a reported 9 million unique viewers per month (Google Ad Planner), 11.9 million (comScore), or 19 million, as the site itself reported from their logs.

But the question of bias remains a tough call, given that Google's measurement methods are shrouded in mystery. Some worry Google's means of measurement relies too heavily on its Toolbar data, which would bias data toward Google sites.

Google defended its Ad Planner data sources, arguing they vary from aggregated search, opt-in anonymous Analytics, opt-in external consumer panels, and other third-party market research. "Google Ad Planner in no way treats AdSense sites differently than non-AdSense sites," a spokesman said.

Off the Google campus, the debate rages on. One company, Startupbin, created a custom Google search site called Google Minus Google, which enables users to run online searches without getting results from Google sites like Knol, Blogger and YouTube.

When Knol went live last month, users became suspicious about Knol pages' capacity to rank well in searches, even though the Wikipedia rival had only existed for days.

Google currently also accounts for 77 percent of all search engine ad spend, receiving $1.10 for every new dollar spent on search.

Responding to this article on August 21, a Google spokesperson wrote:

Google Analytics does not share individual site-level information with Google Ad Planner. These tools gather data from multiple sources … checked against anonymous, aggregate Industry Benchmarking data within Google Analytics. Benchmarking data comes from the setting within Google Analytics that enables customers to opt in to share their data in aggregate (see this blog post: http://analytics.blogspot.com/2008/03/benchmarking-now-available-plus.html).

Anonymous Google Analytics data can be used by Google Ad Planner to calibrate category data and correct for under or over reporting in certain verticals.