Matt Gielen’s Cracking The YouTube Algorithm 2020

Last updated on June 14th, 2024

Check out Matt Gielen’s videos and articles on “Cracking the YouTube Algorithm in 2020”:

• Watch the Video
• Read about it on Tubefilter
• Read the SlideShare presentation

TRANSCRIPT:

Dane Golden:
“So the YouTube algorithm has been deconstructed, it’s been demystified once again for 2020, this is Matt Gielen, we’re going to talk to him about it, coming up. Hey! This is Dane Golden from VidAction.tv. This is the channel where marketers and business owners, just like you, learn about how to make your YouTube channel better. I’m here with Matt Gielen, of Little Monster Media. He again here at Vidcon has deconstructed, what is it? Did I call it reverse engineering? Or reverse?”

Matt Gielen:
“This one was cracking.”

Dane Golden:
“Cracking the YouTube algorithm.”

Matt Gielen:
“The white paper reviews are the… really Dane?”

Dane Golden:
“Yeah, we’ll see if it turns. It turns. Yes, go on. You have to speak, you have to project.”

Matt Gielen:
“The white paper reviews are the reverse engineering ones, the cracking is the big data reports that we do to see what we can pull out about what the algorithm prefers.”

Dane Golden:
“What is an algorithm, and why do I want to know what the it is?”

Matt Gielen:
“An algorithm decides which videos to put in people’s feeds, or to put into suggested videos. It’s basically a choosing mechanism for large platforms.”

Dane Golden:
“A choosing mechanism. So on YouTube, what you’ve been doing this several times over several years, it finds out what’s important and how long people do it, and things like that. So what were some of the big learnings you talked about in your session, and we’ll link to your video you did about it.”

Matt Gielen:
“Sure I think some of the biggest ones are kind of the ideal upload frequency, about three uploads a week.”

Dane Golden:
“Three uploads a week.”

Matt Gielen:
“Or, more than one a day, right? Once you get to, there’s a bit of a dead spot between three a week and one a day, and then once you get to one a day it’s just how many ever videos you can upload on a daily basis. We weren’t able to look at click through rate, mainly because it’s not accessible via API, apparently. So click through rate’s a big one. And the next big one is average view duration. So as it pertains to average view duration, the best sweet spot is still five to eight minutes in an average view duration which will get your video served pretty broadly.”

Dane Golden:
“Hey, it’s Dane back in the studio. Great interview with Matt Gielen of Little Monster Media, talking about his Vidcon presentation on cracking the YouTube algorithm in 2020. I want to go through these slides, step by step, and really dive deep into what he’s talking about. I’ve spent a lot of research doing this, and hopefully I got it right, only Matt will know!”

Okay with Matt Gielen’s cracking the YouTube algorithm in 2020, he runs Little Monster Media, and he did this in conjunction with TubeBuddy, they anonymized the data of these many thousands of videos. 200,000 videos to be exact, across 20,000 channels. They did not include channels with less than 10 videos, and they also excluded channels that had one or two outliers which you might call a viral video, that can really skew the data. So he looked at channels, apparently, it says here, between mostly 10 and 50,000 subscribers, 50,000 and 100,000, 100,000 and 250,000 and so on. The area above a million is the smallest and that’s, knowing Matt, he probably would have loved to have these all the same amount of channels, but there are just fewer channels that have more than a million subscribers, so those are harder to track.

VIEWS SUBSCRIBER CATEGORY

Okay, what is views by sub category? Now, the sub category, he decided, was a group of channels that had between certain numbers of subscribers. So, again this 500,000 for instance, would be between 251,000 and 500,000. So, what this is showing, day one, day two, day seven, day 30, those are the number of days after the video is published. And what they’re seeing is, how much growth is there as the video progresses? Well one of the things that they’ll show is that the more views you get initially, the more views you will get in the end, after the lifetime of the video has gone. Well what we’re seeing here also if you notice the difference, now we don’t have the exact numbers, but if you notice the ratio, the largest channels seem to be doing much better later in life. Whereas the smaller channels, there seems to be less of a difference between day seven and day 30. As you go on, there seems to be a much greater difference between day seven and day 30. Meaning, the best videos, the best channels are being favored because probably they’re being watched longer, thus they’re resurfacing in the feed for people to watch.

AVERAGE LENGTH BY SUBSCRIBER CATEGORY

Well, what about how long these videos are? Now, the exact length, I think Matt will tell you, has nothing to do with anything. It’s how long they watch of an exact length. And it looks like most of the best channels are doing more than 10 minute videos, and the longer channels are really up more towards 12 minutes, the larger channels.

AVERAGE LENGTH BY MONTHLY VIEW RANGES

Okay now he did, I don’t know why this number here is 1,000 and 10,000, most of what he was talking about was towards the 50,000 range, but we’ll just say that here in the 50,000 range, they are edging up towards 10 minutes per video, and then what happens when you get into these larger and larger channels? Well, you are getting towards longer and longer videos. So why is that? Do they just think that the act of doing a longer video is getting them more views? Matt doesn’t think so. He believes that 10 minutes and 30 seconds may seem like a shorter duration than nine minutes and 59 seconds.

AVERAGE VIEW DURATION BY SUBSCRIBER CATEGORY

Okay, average view duration by sub category, again the sub category is the amount of subscribers. So what he’s saying is the average view duration by the channels that have the most subscribers is actually the longest. So when you have an average view duration of five minutes, and remember, these channels were getting videos that are around 11 minutes, means its roughly about 50 percent average view duration. Now you’re starting to see people very, very engaged for a long time. And YouTube loves people to stay on longer.

AVERAGE VIEW DURATION BY MONTHLY VIEW RANGES

He did something by monthly view ranges. This is how much views per month. Again, let’s just say over four minutes, once you start getting people to watch a video for that long, you’re doing really well.

PERCENTAGE VIEWED BY SUB CATEGORY

Percentage viewed by sub category, so he emphasized that this was not as important because a short video and a long video could have very different watch times, but average view duration could of been 50 percent, which means one minute on one video, and 20 minutes on another video. That being said, the people that can make somebody watch a given video for longer, remember these people towards the higher end were having longer videos, but they were also getting people to watch longer. So up here, we’re edging up towards 40 percent of the video.

NUMBER OF CREATOR SUGGESTED BY SUBSCRIBER CATEGORY

All right, so what does this mean? Number of creator suggested by sub category? All right, what does creator suggested mean? So creator suggested is, how many of those videos of the same creator are showing up on the watch page along the right hand side, or below on mobile. So on desktop, it’s on the right hand side of the page, you know where those suggested videos are, it’s the videos that you see while you’re watching another video. Now what he’s saying is that the channels that have more subscribers are getting more suggested videos on average. So, I think there’s a possibility without clicking more, of 19 possible creator suggested videos. So, you’re edging up here towards nine or 10. And what that means is you have a much higher likelihood of people watching more of your videos. Suggested is the number one way of getting more traffic. So how do you get in the suggested? Well these people have found the formula, and that’s probably why their channels are getting such higher subscribers. It’s because they’re getting view after view, from the same viewer.

CHARACTERS OF TAGS PER VIDEO BY SUBSCRIBER CATEGORY

All right, characters of tags. A lot of people are saying tags really don’t matter. YouTube is saying it doesn’t matter. It may matter more in the first few days, or four hours before YouTube uses watch time to figure out what your channel is about. My feeling is this is probably more correlated to just people with larger channels, knowing how to use all the features, not necessarily that each of those features are weighted equally. So they’re using it, but really don’t focus on the tags. It’s really not going to be helpful for you. However those key words that you might be putting in the tags, if you find those are valuable topics, those are helpful for guiding your topics.

DESCRIPTION LENGTH PER VIDEO BY SUBSCRIBER CATEGORY

Description length, okay, what this is showing is that the longer channels are doing more subscription length. Again, that may be a correlation, but I think it’s probably a causation because most of their data, most of YouTube’s data comes from what is in your text, and the more text you give them, the more information they have about what’s in your video, but it’s probably also correlated to just people knowing how to use all of the features, and knowing that you can put in whatever, 5,000 characters or something.

PERCENTAGE OF VIEWERSHIP FROM SEARCH BY SUBSCRIBER CATEGORY

Okay, percent of viewership from search by sub category. All right you’ll hear a lot that search is not the main driver of traffic on YouTube, and people come from Google say, let’s rank and search. However my belief is that search is a very good indicator of what people are identifying your channel for, and also I think that search helps understand what your channel is about, and your videos are about, and those do influence suggested in a way, but as much as watch time. And Matt Gielen said this himself, search does help influence, but not as much as your actual viewing behaviors. So your viewing behaviors will influence much more, how much will you show up in suggested videos, which is again the number one traffic source. Now a channel like mine that is small, more of my traffic is going to come from search because people are not going to know it as well, the algorithm doesn’t know it as well, thus more is going to come from search. But as you go on, and the algorithm really loves you, and you’re making content that is really resonating, then you’re going to find you’re getting more traffic from elsewhere, including from suggested.

AVERAGE MONTHLY UPLOADS BY MONTHLY VIEW RANGES

Average monthly uploads by monthly view ranges, okay, basically let’s just shorten this here. What he said was, the channels that were uploading three times a week, 12 times a month, or more, were doing better. That being said, a lot of these channels with very, very high subscribers, are media channels, they are to entertain. Most of who I work with are business channels and I’m sorry, you’re just not going to get a million subscribers.

AVERAGE MONTHLY VIEWS BY VOLUME OF UPLOADS

Average monthly views by volume of uploads, again that 12 sweet spot, he really believes, is the best, and then once you start doing, you know, one a day, okay that’s great. But 12 a month, that’s three a week right? If you can do that, you’re going to do just fine. But remember, the people that he’s studying, for the most part, this is weighted towards people who already do good channels. And just because you make an upload does not mean people are going to watch it, so make it a good video. It’s better to have a good video than a frequency.

TRAFFIC SOURCE AS A PERCENTAGE BY SUBSCRIBER CATEGORY

Okay, traffic source as a percent of sub category. All right, what is this saying here? Let’s take a look at search here, and we can see that search is less than suggested and browse. All right, we know what search is, that’s traffic that comes from YouTube search, not Google search. Now, suggested, what is that? That’s what comes in the right hand column, or right below on mobile. What is browse? Browse is basically what shows up on the home page, or mostly on your phone when you just open the YouTube app. Well, why is this going down as the channels get larger, but suggested is going up? Well, first of all, there’s only 100 percent, so they have to exchange somewhere. These are usually the top three percentages, and for smaller channels again, YouTube doesn’t know you as well, and they are not suggesting you as much. However he said elsewhere that browse is a lot more important in the first couple of days, suggested is what’s more important long term. And when you get that suggested long term, remember we saw those large channels doing much, much better later on.

PERCENT OF VIEWERSHIP FROM OTHER SOURCES BY SUBSCRIBER CATEGORY

Okay percent of viewers from the other sources. Now these are other important sources. Now I want to talk about an area of disagreement I have with Matt, and he says notifications are usually never higher than two percent, so stop asking for them. That’s something I disagree with because notifications are of your most loyal users and he has said, of your most loyal users, that’s one of the most important ways of YouTube determining how good your individual video’s going to be initially. So assuming that these are loyal viewers, this is going to be very, very important, because this is the test case. So there’s no reason why you should stop asking people to get notified.

PERCENT OF VIEWERSHIP FROM OHER SOURCES BY SUBSCRIBER CATEGORY

What about external? What is external? So that could be Google search, which is a subset of external. It could be anything embedded in a website, stuff like that. I think e-mail is included in that as well, so if someone e-mails you a link, or if you send out a link in your newsletter. Playlists, playlists are a small area, but don’t underestimate them, they are very important. In fact, you can see how important they are as the channel gets larger, so do your playlists, that’s one thing that channels do not do that’s a big loser for them because they’re not doing it.

TRAFFIC SOURCES BY MONTHLY VIEW RANGES

All right, traffic sources by monthly view ranges. What the heck does this mean? All right, so average browse percentage, average suggested percentage, and average search percentage. What this is saying, it sort of repeats what we said before, is that the larger your channel gets, which is over here, the smaller search is important, the less browse is important, and the more suggested is important. But these three elements continue to be the most important.

VIDEO DATA

Now on video data, he didn’t include anything about live streams, advertised videos, music videos, videos with less than 100 first day views, or 1,000 seven day views.

COUNT OF VIDEOS BY CATEGORY

Let’s take a look at how he did the study by category. He studied a lot more entertainment, and people in blogs, videos, and a lot less of pats, and non-profits and travel. I think that’s probably just based on the data he got. Category’s not too important.

COUNT OF VIDEOS BY LENGTH

Now let’s look at the count of videos by length versus the number of videos. A lot more of his videos were way over 20 minutes that he studied. Now, let’s take a look actually, this sort of harkens back to what we were talking about before, there’s this peak here at 11 minutes, and there’s a peak at over 20 minutes. Why is this? You can do more ads over 10 minutes. And we also saw that a lot of videos, once they got over 10 minutes, they were getting more watch time.

COUNT OF VIDEOS PER MONTH

All right so this was just based on how many videos he was studying, videos per month.

CREATORS SUGGESTED VIDEOS PER VIEWS

Okay, your own suggested videos versus views. Okay basically what this is saying is that the more views a video has, the more suggested videos it’s going to get from the same channel on that watch page, and that’s just going to go up over time.

VIEWERSHIP BY DAY OF THE WEEK UPLOADED

Okay, viewership by day of the week. Basically day of the week doesn’t matter, day of the week only matters depending on what type of channel you are on. If you’ve got a channel catered towards kids, the weekend and Fridays are going to do better, if it’s catered towards people at work, the weekdays are going to do better probably Monday through Thursday.

DOES THE VIDEO HAVE CUSTOM THUMBNAILS?

Custom thumbnail, hey you know what a custom thumbnail is. That’s just whenever you change the thumbnail, and you customize is to get more clicks. What is this telling us? This is telling us how important it is, because from the beginning to the end, it will always get more views if you’ve done a custom thumbnail, so don’t ignore the custom thumbnail, it is important. Thumbnails are extremely, extremely, extremely important.

DOES THE VIDEO HAVE CAPTIONS?

What about captions? Okay, he’s talking about manual captions here. That’s when you work with someone like Rev.com, link in the description to my affiliate, if you get someone to manually caption your videos. Now YouTube is likely going to caption the video to begin with with automated captions. But the problem is, is that some of the key words that automation is going to get wrong, often the new words, right? Which is the one you want to emphasize. So look at this, look at how much better over time these videos are going to do with captions. Looks to me, just by eyeballing it, about 40 percent better if they have captions. Does that make you want to do captions now? I think it does.

TITLE LENGTH VS. VIEWS

Title length versus views, basically what this is saying here is that the shorter the title, these are videos between about two to seven words, three to seven words, are going to do the best. Three to six words are going to do the best. Doesn’t mean you can’t do well with a long title, but what it’s really saying is that short titles really, really, really, really matter.

TAG COUNT VS. VIEWS

Tag count versus views, again, I think this is more of a correlation. Although it does go up at the end. So what does that really mean? It probably means that the people who know what they’re doing are also just taking a little bit of time and doing a lot of tags, but it’s probably because they already got to where they were going and got to be good because they can make good videos and they’re also using all the features.

LIKES VS. VIEWS

Likes versus views, well basically likes don’t matter for a given video, but they will matter for future videos. If you like a video by one creator, that might influence what YouTube sends you later. It’s probably going to be more influenced by the watch time itself, did you watch it a long time versus the like? But it’ll be a signal, we can just skip this one.

VIEWS VS. COMMENTS

And the same with comments, not really terribly important. Other than, if somebody comments, and you comment back, and they return, that’s two views on that video for them because they’re watching it as they’re commenting, so nothing wrong with getting comments, but not a factor in ranking.

LIKES VS DISLIKES

Okay likes versus dislikes. I think it doesn’t really, this is more a correlation, it’s watch time that’s going to matter. Same with this slide.

LENGTH AND AVERAGE VIEW DURATION (1,2,7,30,60,90 DAYS)

Length and average view duration. All right, so the length of the videos, the overall duration of the video, and the average view duration is how much of that video someone’s going to watch. So here’s the length here, and again you’re seeing this jump at the 11 minute area, largely because people can put more ads on there, but also because maybe it seems a little bit shorter to have something that’s 10 minutes and a few seconds versus nine minutes or eight minutes. But we’re also seeing this jump here at six minutes. Now, it’s not necessarily a better jump in the short period. Now, this is a little bit difficult to see, but if you look quickly here at 11 minutes, this is at 40,000 here after two days, and then for six minutes it’s probably at about 30,000, right? So after two days, the long video is going to get more traction, but after this longer period, three months, the videos that really do well are going to be a little bit shorter, and if you think about that logically, it’s going to make sense, because you’re more willing to invest time in something that’s shorter if it’s old than something that’s longer. Most likely. Because you think it might be dated and why should you invest all this time in it.

DAY 1-7 AVERAGE VIEWS VS. SECONDS

And we’re seeing that same, so this is 600 seconds which is 10 minutes, and we’re seeing that same bump, is that the better channels are going a little bit longer and probably getting more watch time as well. Just actually having a longer video is not going to help.

AVERAGE VIEW DURATION VS. VIEWS

Okay, what about this slide, average view duration versus views. So remember, average view duration is the amount of time that people are watching, not the amount of time that the video could go on, potentially. So remember, the sort of sweet spot was around 11 minutes, but people are not watching that. Remember we saw previously that the average view duration overall tended to be more towards 40 percent. And if you think about it, let’s say 10 minutes and we’ll take 40 percent of that, that’s more like 4 minutes, and this is the average view duration here, well this looks pretty close to a peak here, some are larger and some are smaller, but the six minute area looks like it’s the largest over time. So that means if you have, essentially, a 10 or 11 minute video, you want someone to be watching it at least for about six minutes. This is the same as the previous slide.

AVERAGE VIEW DURATION VS. VIEWS DAY 1-7

Okay, and this is where we’re getting towards, it’s just another way of looking at what we were seeing before, 360 divided by six is six minutes, so that was the average view duration. If you can get people watching for about six minutes, you’re doing really, really well. If you can do that consistently, you’re going to be among the most popular channels in your category.

THE LATENT CROSS

Okay, what is the “Latent Cross”? This is from YouTube engineers. And basically, what the “Latent Cross” means is that YouTube is getting really good at understanding both the time of day and the device you are using as to what type of content they should serve you. If you’re on mobile, they’re going to serve you shorter content. If you’re on desktop, they’re going to serve you longer content. Let’s say you take a commuter train to work, and YouTube knows that you take a commuter train and thus you are likely to want to watch shorter content because that’s what your experience has demonstrated in their algorithm. That’s what latent cross means, I think.

PERCENTAGE VIEWED VS. VIEWS DAY 1-90

All right, percentage viewed. Now we talked about this before. Now, percentage viewed, there’s this sweet spot here of about 50 percent, you basically want to keep people watching, essentially. The longer you can get people watching, and Tim Schmoyer and others have said this, but if you can get them watching all the way to the end, or most people to watch all the way to the end, then you will see a dramatic increase. And part of that is that if people watch all the way to the end, then you’re going to see someone clicking on your end screen and they’ll watch another one of your videos which means YouTube will promote the previous video, the one that you’re on here, the one being marked here. So the more people you can get watching for longer periods, the better off you are. If you’re down here, you’re not doing very well. Similar here, just keep watching. Also I think this data probably doesn’t have enough data to measure this but I’ll bet if you’re watching something all the way through, that would be probably even better. But he probably doesn’t have enough data for that.

SUGGESTED AND BROWSE

DAY 1 BROWSE PERCENTAGE OF DAY 1 VIEWS

Suggested and browse, these are the two most important ways of tracking data. All right, so what is percentage of day one views from browse? All right, okay I don’t know what this means.

DAY 2 BROWSE PERCENTAGE OF DAY 2 VIEWS

I’ll sum up what he meant when he was talking about browse. And when he was saying basically, as I understood it, if you get a lot of traffic on day one and day two from browse, you’re doing really well. Overtime you’re going to do best if your traffic is coming from suggested videos, but neither of those is something you can very much change. You just have to get people clicking, and you have to get them watching longer. But basically, you can’t have both browse and suggested at the same time. So what this is saying is that, if one is larger than the other is smaller.

SUGGESTED AS PERCENTAGE OF DAY 1 VIEWS

So the more you get suggested, the better you’re going to do long term.

AVERAGE VIEW DURATION VS SUGGESTED VIDEO VIEWS AS A PERCENTAGE OF 1ST 7 DAY VIEWS

Average view duration versus suggested videos. Okay, remember when we said the average view duration, the sweet spot of getting most views is around here? Well basically, it’s saying that the longer video is, the more it’s going to be in suggested videos early on.

AVERAGE VIEW DURATION VS. SUGGESTED AND BROWS VIDEO VIEWS AS A PERCENTAGE OF 1ST 8 DAY VIEWS

But I don’t think that’s what you want, I think you want it in browse features early on. So browse is the brick colored thing here, and you can see early on, it’s going to do a lot better, particularly for channels in that six minute sweet spot, right? Six minutes, they’re going to get the most browse features. And now it’s what you want, and then overtime browse features is not going to be as important and suggested videos is going to be more important.

AVERAGE VIEW DURATION VS. VIEWS BY BROWSE AND SUGGESTED

Okay, same thing. Do more, get longer watch time, get at least six minutes. You’re going to do just fine. I don’t really need to go into this.

PRACTICAL DIVERSIFIED RECOMMENDATIONS ON YOUTUBE WITH DETERMINANTAL POINT PROCESS (THE DPP OPTIMIZER)

All right, okay, so what is the DPP? The “Practical Diversified Recommendations on YouTube with Determinantal Point Process.” Basically what this means is that YouTube realized that by showing you the same kind of video, even if they thought that you only wanted a certain kind of video, it was actually better to show you fewer of those than more of those. And show you a variety, even if they didn’t think you were going to click on this other video, give you another one. Like you’re trying to look up videos that will increase your watch time, so you keep watching more videos that increase your watch time even if the last 10 videos were about how I can increase my watch time, YouTube’s still going to give you a cooking video, and a silly comedy video as a choice because they realize that they don’t know the point at which you’re going to want to change your mind. And they can keep you watching when you decide to change your mind. When you’re like I’ve had enough work, now I want some play. Or I’ve had enough on this topic, I want something on this topic. So the idea is give you a little bit more variety, even if the likelihood is small of you clicking off that one continuous topic viewing session, because the goal is to keep you on watching, not to keep you watching a given topic, they don’t care about what topic you watch, they just care about whether you’re on or not.

Dane Golden:
“Okay, now where can people find out more about this?”

Matt Gielen:
“People can find out more on the Little Monster Media Co. YouTube channel, on our website, LittleMonsterMediaCo.com, and on TubeFilter where we release papers that we write.”

Dane Golden:
“Okay, that’s Matt Gielen, check out his YouTube channel, Little Monster Media, and subscribe to mine. Watch the next video, you do that and you’ll be Golden.”