The answer to the oft-asked question of what people want to see on the Internet is pretty much the metaphorical Holy Grail for Web publishing. If we knew the answer to that, then you’d likely see a lot more of it and a lot less of everything else – even if only for getting hits and not the more altruistic aim of giving the people what they want. However, one website may have worked out a way to predict what stories are going to be hits before it happens – and if it proves to be a consistent predictor of tastes, the Internet as we know it may be able to change.
Adweek reports that a software built by the team behind Conde Nast’s Ars Technica site can track current pageviews and use that data to extrapolate how successful stories are likely to become. “Within two days [of being installed] we found it was working really well,” Ars Technica editor-in-chief Ken Fisher says. “We were identifying within an hour stories that would go on to do 900,000 views. And these were not pieces you’d hear by title and think, ‘That’s going to be killer.’ One was titled, Quantum Networks May Be More Realistic Than We Thought.”
The technology is currently dubbed “the accelerator” (or, alternatively, “the watercooler program,” presumably after the hypothetical water cooler that the story will be discussed around should it prove to be appropriately “sticky”), and tracks not just raw pageview data in real time, but also where the traffic is coming from. “We can know after 20 or 30 minutes if organic traffic will be huge on a piece,” Fisher says.
The site continues to boast that the program – which is said to be still new and in the development stages – has already found a “more than 95 percent” success rate in predicting which stories are going to be successful on the site. Ars Technica publisher Howard Mittman said that, should the accelerator continue to perform as well as it has done so far, “our hope is [that] we can roll this technology across other sites and help Conde Nast continue to grow.”
While the technology as currently described can only predict behaviors of already-written and posted stories, if this model proves to be accurate on an ongoing basis – especially to 95 percent – then it can only be a matter of time before the arrival of a predictive model that can suggest whether or not a story will be successful from a one-sentence pitch alone. Could this mean that Internet publishing would become more focused on the big stories only, or that we’ll see more niche publishing in the future?