YouTube: Good Luck Trying to Game Our Algorithm

YouTube: Good Luck Trying to Game Our Algorithm

It takes more than one million lines of code, says Christos Goodrow from YouTube, to figure out what to recommend on the site. The average YouTuber refers to this as the algorithm and it is constantly changing. A significant change to parts of this code can make or break a YouTube channel, yet the algorithm is often a mystery to creators. In a recent interview with Computerphile, Christos shed some light on just how the current code works.

The reasons for the code being such a secret are very simple. There is a lot of money and fame to be had through YouTube and knowing the code would allow somebody to easily game the system. Many video creators in the past did just that once they figured out how the system rewarded certain behavior. One of these instances was the “reply girls” phenomenon. Back when view count was more highly rewarded, they gamed the system to the tune of millions of views and subscribers by posting their videos as responses to popular or trending videos. This, combined with provocative titles or thumbnails allowed them to garner lots of fame and infamy. Another way to game the system that is still going on, is paying for views in order to keep a video more highly ranked in search results. Both of these methods are discouraged by YouTube and they have taken measures to curb their effectiveness.

“I don’t want to lock us into a particular way of doing things . . . 5 years ago ranking the search results by view count may have been a great thing to do. And that was because at that time, view count was one of the best signals that you could find for what are the quality of the video.”

More important than just the bits of code are the purpose behind them. This code is what drives YouTube and serves as the gatekeeper and benchmark for what type of content the site encourages creators to make. One of the most important aspects of the algorithm is that it must be based on honest results. When a robot or person is paid to increase the views of a video or a viewer is duped into watching a video they aren’t genuinely interested in, the results can be skewed.

Watch Time is Everything

View count doesn’t weigh nearly as much into the formula any more, says Christos, mostly because viewers would never see any new videos that way. One of the main signals for video success today, which has been used the last couple years, is watch time. When a person makes a search query on YouTube, the longer they watch the resulting video, the more favorably it will appear in similar search results the next time.

As any creator can tell you, the YouTube algorithm can be quite volatile. Christos and YouTube realize there are flaws in the code and they are constantly working to improve them. At any given time he says there are at least ten changes being worked on that could take anywhere from a few weeks to a few months to implement. He says that, “I think we’re open about the instances where we know it’s not working and we think it could be better”.

In the end there is one great piece of advice for video creators. Make compelling content and be honest about what it is with the viewers. Gaming the system can pay dividends in the short run, but if a creator is going to be making content for the long haul, eventually the algorithm will catch up with them.


Don't Miss Out - Join Our VIP Video Marketing Community!
Get daily online video tips and trends via email!


About the Author -
Andy Smith is the host of ReelSEO's "Behind the Reel" show. He has been exploring what makes YouTube and its creators tick, on his own YouTube channel, since 2010. View All Posts By -

What do you think? ▼
  • http://www.youtube.com/user/Pokeflirt Michael Vera

    Awesome write up, thanks my man. I wonder if "Watch Time" will ever by overtaken by some other factor in the future.....maybe social signals? Like shares or comments. Who knows hmm...

    • KH

      You can bet that shares, comments and "likes" are most certainly being used together with watch time and view counts to rank videos. I would bet that they exist mainly to feed data into the algorithm!