Measuring internet video, like math, is hard.

Witness the wildly divergent numbers for hulu coming from competing measurement services Nielsen Online and Comscore: as reported a few days ago in the New York Times, Nielsen counted 8.9 million unique visitors while Comscore counted well over four times that number: 42 million.

Understandably, hulu takes issue with the lower Neilsen numbers.

The challenge of getting accurate viewership numbers in the decentralized on-demand world of internet video is at least an order of magnitude more difficult than doing so for linear broadcast video, but both companies are hard at work on the problem.  Neilsen has just increased the size of its “Netview” representative panel from 28,000 to 200,000  subjects.  While Comscore has a similar “Media Metrix” panel, Nielsen also utilizes a census model (much like Google Analytics, this  system requires that a partner content publisher embed a small amount of javascript code within their web content that ‘phones home’ to the Neilsen servers).

Despite their best efforts, though, in the case of hulu’s March numbers at least one of these two companies is very, very wrong.  And will be important to get it right (or less wrong, at least) soon – because over the next few years as internet video trends away from long-tail user-generated content (UGC) viewed alone on a computer to a more television-like experience involving premium content, advertising is going play much more of a role in the business model than it has to date.

Against this backdrop, it was interesting to listen in on a scribemedia.org web discussion yesterday on just this topic: online video media measurement.  Participating were Todd Juenger of TiVo and Jon Gibs of Nielsen Online.  While both acknowledged the disruptive impact internet-enabled CE hardware will have, Mr. Juenger suggested that such hardware could also present a possible solution to the measurement problem: several hundred thousand  TiVo subscribers have already opted in to a program in which their current viewing of time-shifted broadcast content is tracked electronically.  Perhaps as the next generation of network-enabled set-top devices become more common, accurate metrics could be similarly obtained via the hardware on the client-side, thus eliminating the need for server-side scripts and/or panel-based statistical estimates.  The carrot that enticed so many to volunteer away their privacy?  A sweepstakes prize of precisely one new TiVo DVR (talk about bang-for-the-buck: give away one device, get extremely granular hard data on hundreds of thousands of customers).

You would think users would be more reluctant to have their internet viewing habits constantly tracked, but if the case of TiVo is any indication, maybe not…