Game Theory's Decoding of the YouTube Algorithm

Dewmonic Abyss

Super Poster
If you haven't seen MatPat's recent video: Game Theory: Beyond Fidget Spinners – How to Create a YouTube Trend he actually does a very thorough analysis of the YouTube algorithm and brings up a lot of very interesting concepts. Of course, you all know the generic stuff 'Have detailed descriptions, good search engine optimization, engaging thumbnails.' But it turns out there's actually a way YouTube might be thoroughly judging videos before recommend them to you. And one of these has to do with what you've searched for recently and what you've spent the most watch time on. Which is exactly why a lot of memes can become memes. You get recommended ONE, click on it, laugh, see another pop up in your reccomended, click on that one, and before you know it, you're only being recommended We Are Number One and the Bee Movie but distorted beyond belief in slow motion and every time they say Bee it's the SOMEBODY scene from All Star. And even demographics. By trying to appeal to everyone, rather than a specific group, you're more likely to not be noticed at all.
Another thing he brought up was that YouTube wants really fresh content. They want to focus on what's trending and what's relevant. So, if you plan on having a retro gaming channel, not only are you less likely to get views because are searching for new games, but YouTube isn't even going to think twice about promoting you, unless someone is only watching retro games. The way he described it was pretty well, "like YouTube is going through my videos and squeezing them to judge them... like mangoes or other ripening fruit."

And according to this, explicit feedback (such as number of Likes) actually has very little to do with how you rank on YouTube. So asking people for likes actually serves no purpose. YouTube wants you to make videos that are new, long, trending, and target a specific demographic. And consistency is even more important than we give it credit for. He's cited that many music channels have suffered because they uploaded behind the scenes vlogs for their music, which people would skip over and wait for the music, and this essentially kills their channels because YouTube thinks people don't think they're interesting! Or even why channels that focus on only FNAF or Minecraft would not get views on other types of games.

And even more, YouTube has a thumbnail algorithim! Did anyone know this? YouTube can apparently recognize if a thumbnail contains a face, a color, or even the style! So, it seems that thumbnails that have prominent faces in them are more likely to be promoted by YouTube. And finally, Google came out with an algorithim that will allow them to gauge whether a video is funny or not, simply because viewer interaction in the comments. If people are saying 'lol' a lot, that video is more likely to be promoted to an audience that appreciates comedy.

This stuff all blows my mind, and I couldn't even begin to cover all of it in this post, so I highly recommend you check out the video I mentioned above. It's not speculation, as this was all taken from accounts of Google employees, which makes sense, because I'd assume they'd want their creators to do well, so YouTube can generate more money. I have mixed thoughts about the algorithim. For one, like I mentioned, retro channels are gonna suffer from this. As well as animation channels, and other content creators that take a while to produce content and risk the chance of making content on irrelevant topics. I'd even suffer from this, as I have a very flexible channel that does both reactions and gameplay, two different genres, but both that fit my personality. What are all your thoughts on this?
 
Awwwww man, this is good and bad I guess. I hate how YouTube can just choose who is favored. I personally think meme vids are entertaining and fun to watch but I would much rather see the algorithm favor people who put out high quality content albeit inconsistently or infrequently. Meme vids can take no time at all to make and rake in views and ads all day long. It hurts my soul.

Thanks so much for this post, I'll have to check out the vid ASAP but your summary is a great run down. Cheerios!
 
It's a great video, lots of good stuff in there. My takeaway from it, with the implementation of machine learning, it is much much harder to both build, and then maintain growth, of a channel. Yt is becoming like an Olympic sport, only a very few climb to the top, and can maintain their rankings for any significant amount of time. It needs a lot of work, forgot about the 40 hour week, more like 80 hour week, and being on the cutting edge with creativity and content development, technical skills, and a flawlessly executed back-end.
 
It makes me wonder if someone could make a video that is total nonsense but hits all the right notes and could be pushed to the top. I guess the decider in that would be if people would sit through such a thing. Make sure to pepper your videos with certain words and images I guess
 
It's a great video, lots of good stuff in there. My takeaway from it, with the implementation of machine learning, it is much much harder to both build, and then maintain growth, of a channel. Yt is becoming like an Olympic sport, only a very few climb to the top, and can maintain their rankings for any significant amount of time. It needs a lot of work, forgot about the 40 hour week, more like 80 hour week, and being on the cutting edge with creativity and content development, technical skills, and a flawlessly executed back-end.

What all do you do when you say working 40 hours? Besides filming and editing, what else can you do to work on your channel? You are right, it is like an Olympic sport.. and I suck at sports haha.. I don't work near 40 hours a week on our channel.. maybe that's why we aren't successful.
 
No more face on my thumbnail except if I draw one so I guess I am sure I'll stay a tiny youtuber then haha especially since my videos tend to be short. There was a guy called 3 minutes Japan saddly for him his concept prevented him to grow I guess he is not a tiny youtuber but I could see him be bigger.
 
No more face on my thumbnail except if I draw one so I guess I am sure I'll stay a tiny youtuber then haha especially since my videos tend to be short. There was a guy called 3 minutes Japan saddly for him his concept prevented him to grow I guess he is not a tiny youtuber but I could see him be bigger.
I don't like crazy face thumbnails too!

I think you can still succeed with short videos but at first it's not helpful. I'd like to make shorter vids because I can get more put out but I'm sticking with long form.

I think I know who you mean. Could there be more annoying of a channel?
 
Good video... along with the recent Matt Gielen article on Tubefilter.

It's important to remember that many of the results from experiments that people do or observations they make are correlational. We don't know the exact mechanisms, nor what hidden variables exist. What we do know from Youtube's own paper is that they keep track of hundreds of variables, and their interactions are highly nonlinear.

That said, my own observations and intuition tell me that consistent quality and high click-through-rate (via good thumbnails) are among the most crucial factors for getting promoted more. If you have content that works well with trends, then keeping on top of the latest trends is also super important.

Improving your own content to make it engaging (on a consistent basis) is one of the hardest parts of making videos. We can be blind to the quality of our own videos because we are making them ourselves, or feature people that we care about.

Anyway, just my 2 cents!
 
Agree with all you said but when you say "high quality content" in YouTube context you always must keep in mind that this is a toddlers/small kids type of "high quality content".

In machine learning terms - the video "quality" content is tested against some very young and underdeveloped brains. Keep this in mind as it will influence enormously the final result of the experiment.

Just my 2 cents on the subject. :)
 
Back
Top