YouTube mysteriously, and without an explanation, changed its recommendation algorithm last month to penalize videos that were garnering a high volume of clicks but weren’t keeping the attention of its viewers. The move may have indicated YouTube’s favoritism for original programming. But during an Ad:Tech panel this week, Laura Lee, Director/Head of Entertainment East Content Partnerships revealed that the decision to implement the Time Watched algorithm came after brainstorming ways to systematically prevent YouTube content creators from gaming its search engine.
The “gaming” she’s referring to is the use of thumbnails that misleadingly advertise videos. Content creators can take any single frame from their video and feature it as a preview thumbnail. These thumbnails proved to be a critical component in luring viewers to click on the video. Any YouTuber can tell you that it’s becoming increasingly difficult to have their content discovered on the platform, so creators will take any edge that they can get.
How it works, according to Lee, is that YouTubers will add just five seconds of irrelevant, often lewd or graphic, content for the sole purpose of using one frame from the content as a thumbnail.
“We talked to our product team and did a lot of A/B testing, and we felt that this was a bad user experience on YouTube and the users might not come back,” says Lee. “Part of our tinkering was that we didn’t reward those gaming the system.”
Even some of the upper echelon YouTubers weren’t immune to the update, and felt the initial impact. “All of a sudden some partners were like, ‘hey my view count isn’t as great as it used to be, what’s up with that?’ And not always but sometimes they were gaming the system,” recalls Lee.
She makes sure to point out that this change doesn’t necessarily favor feature length content and it shouldn’t be a logical extension of what content creators do. “Time spent for us could mean someone coming to watch five two minute clips for a total of 10 minutes. Or it could mean watching Machinima’s 19 minute Battlestar Galactica too.”
“We just want to make sure that we’re doing everything we can to surface content that’s relevant and very meaningful to the niche user,” Lee adds.
Since YouTube is able to figure out that you’ve watched five two minute videos rather than one 10 minute video, there’s probably a machine learning algorithm behind its search engine. When I asked Lee before the conference if YouTube could confirm that it was tracking its user’s viewing habits for the purpose of recommending targeted content, she neither confirmed nor denied my speculation.
But let’s entertain the theory for a second: It doesn’t take a fortune teller to see that the natural progression for YouTube’s search algorithm will evolve, if it hasn’t already, into an engine for suggesting targeted results. In other words, based on individual viewing habits, soon the same query searched by two different people may suggest two different, but tailored results.
With YouTube maintaining its position as the second largest search engine on the Web, if targeted results become a reality it could have a far-reaching impact for marketers and content creators alike.