Media, search engines, advertisers and social networks have been tracking what you click since the birth of the web, but this measurement yields an incomplete picture of what you’re actually doing when you browse. What marketers, advertisers and the analytics junkies who serve them would really like to know is what you’re thinking, and the gold standard for determining that is gaze-tracking studies, which can only be conducted in the laboratory.
The key to their innovation is that they track where your cursor is at any given moment. It turns out there’s a high correlation between what we look at on webpages, especially search results, and where we place our mouse cursor. Even more intriguing: tracking cursor position provides information about the relevance of search results that is richer than simple click data.
The researchers’ trial of their script was conducted only on searches coming from people who worked at Microsoft, so it’s not clear if anyone has yet implemented this on a publicly-accessible website. When they do, the implications could be profound.
For example, much advertising on the web is impossible to value except through the blunt instrument of click-throughs. This makes it difficult to measure the effectiveness of banners, brand advertising and other forms of sponsorship that are about building mindshare rather than inducing users to click. Tracking where a user’s cursor hovers – in other words, where their gaze falls – could allow media buyers, for the first time ever, to evaluate who is really looking at their ads, and for how long. And we’re not talking about a sampling exercise, in which they evaluate a few users and extrapolate a larger trend; we’re talking about measuring the behavior of every user.
The code that makes this analytics technique possible is so lightweight there’s no reason it couldn’t be implemented as a standard part of any analytics package. By recording only events in which a user’s cursor was still for more than 40 milliseconds, the inventors of the technique were able to reduce the stream of recorded events to 2 or 3 kilobytes of information that could be sent off when a user navigated away from the page. Recording these pauses, rather than a continuous stream of cursor information, yielded a good approximation of where the user’s cursor had been.