In the chaotic world of modern marketing, the Internet is touted as the best medium for determining the effectiveness of ads.

So why is there so much controversy over a measurement as simple as how many people visit a Web site?

The independent companies that measure online traffic have been accused of undercounting minority Web surfers, overestimating visitors by more than double, and ignoring Internet users at work.

In an example of the sharp disparity that sometime arises, in September three outside groups weighed in on how many visitors the San Francisco Chronicle's Web site, www.sfgate.com, received. ComScore Networks said 2.4 million, Nielsen/NetRatings counted 3.7 million and the Audit Bureau of Circulation came up with 7.6 million.

The dispute isn't just academic. Knowing the size of Internet audiences more precisely is important as advertisers try to decide where to allocate their money. They spent $16.9 billion in online ads in 2006, up 35 percent from the year before.

"Measurability is going to become more important," said Dave Martin, director of interactive media for Ignited Minds, an ad agency in Marina del Rey, Calif. "It would be great not to have to try and guess what traffic levels are."

That's why the two biggest companies measuring Web traffic, Nielsen/NetRatings and Comscore, have agreed to independent auditing of their numbers.

Last week, more than seven years after they first started measuring Web traffic, Nielsen and ComScore said they were on track to finish extensive audits by the Media Ratings Council, which also monitors print, radio and TV, within about a year.

AUDITS PREFERRED

Nearly 70 percent of advertisers said they would prefer to advertise on a site whose numbers had been audited by a third party, said Neal Lulofs, a spokesman with the Audit Bureau of Circulations. Around 84 percent of advertisers said having this verification would become increasingly important.

Right now, there are hundreds of companies out there that count site visitors. But their statistics can vary by the millions.

The two largest, Nielsen/Net Ratings and ComScore, have received most of the heat. The two companies use panels to measure an audience, which count eyeballs similar to the way Nielsen measures TV viewers: They select a sample of people, use software to track what they do online and then use those numbers to extrapolate traffic patterns for the whole Internet.

"There are enormous questions about their reliability," said Benjamin Schachter, an analyst with UBS Investment Research.

One of the research companies' biggest critics has been the Interactive Advertising Bureau, a trade group that tries to increase the amount of advertising money spent online. In an open letter in April, the group charged that the panels "undercount or ignore the diverse populations that are the future of consumer marketing."

"As advertisers are spending more and more money online, they deserve to have confidence in the numbers they're using to make decisions about where to put their money," Sheryl Draizen, senior vice president of the Interactive Advertising Bureau, said in an interview.

Nielsen and ComScore said they had plans in the works to be audited. But they cautioned that the audits in fact might show that the company's methods of using panels to track Web data is in fact the most accurate way — and that the numbers aren't as high as publishers say they are.

Manish Bhatia, executive vice president of Nielsen/Net Ratings, said panel data was the most accurate form of tracking Web-surfing patterns. The alternative is using information collected by a publisher's computer servers to track the people coming to the Web site. Bhatia said that method leads to inflated numbers, because, for example, it might count someone multiple times who logs on from different computers. What's more, he said, computer users often delete the information that is used to track them, so servers count them a few times.

But audited counts from server logs are better than unaudited numbers released by the Web site publishers. Their numbers are almost always higher than panel data. After all, higher numbers are in the publishers' interest — the more visitors a site has, the more it generally can charge advertisers.

"You have to take the publisher data with a grain of salt," said Allen Stern, media director for the San Francisco office of www.Agency.com, which advises clients on buying Web ads.