I Analyzed 93,421 Viral Videos. Here’s What I Learned… (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:00:00) This chart shows the typical video (00:00:02) lengths from the top 100 YouTubers. Now, (00:00:04) in the early days, you can see the top (00:00:06) YouTubers videos usually lasted for (00:00:07) around 5 minutes, which for the record (00:00:10) ladies, is a decent amount of time. But (00:00:11) then things started to trend up a little (00:00:13) bit and recently in 2022, something big (00:00:15) happened. See, Stranger Things season 4 (00:00:18) released with movie length episodes and (00:00:20) the top YouTubers were like, "Bet." And (00:00:22) since then, the top YouTubers typical (00:00:23) video length skyrocketed going all the (00:00:26) way up to 21 minutes. But where does all (00:00:28) this data come from and what does it (00:00:30) mean for small YouTubers? Well, today (00:00:31) I'm going to show you how I downloaded (00:00:33) all the videos and words spoken on the (00:00:35) top 100 YouTubers channels. And we're (00:00:37) going to analyze things like how complex (00:00:39) their words are, how fast they're (00:00:40) spoken, whether they're more positive or (00:00:43) negative, and five other things that (00:00:44) will hopefully help you get more views (00:00:46) than a see-through clothing haul video. (00:00:48) And no, you cannot see my watch history. (00:00:50) Jokes aside, since 2006, the top 100 (00:00:53) YouTubers combined have uploaded 93,421 (00:00:56) videos. In those videos, I found 42,000 (00:00:59) hours worth of content with 286 million (00:01:02) words spoken. Now, obviously, there's no (00:01:04) way I'm going to manually analyze 42,000 (00:01:06) hours of content. I love you guys, but (00:01:07) not that much. So, here's what I did. (00:01:09) First, I made a list of the top 100 most (00:01:11) subscribed YouTubers, excluding brand, (00:01:13) kids, or non-English channels. Then I (00:01:15) paid my developer friend to go and build (00:01:17) custom software that would grab all of (00:01:19) their videos and transcripts. And (00:01:20) finally, I begged my data scientist (00:01:22) mates to help me turn all the data into (00:01:23) enough graphs to resemble the New York (00:01:25) Stock Exchange. And now that we have all (00:01:26) this data, we can do some interesting (00:01:28) analysis to help us understand how it is (00:01:30) the top 100 YouTubers in the world (00:01:32) actually make their videos. But before (00:01:34) we get into that, first I want to (00:01:35) quickly look at how the top 100 (00:01:37) YouTubers actually post their videos. (00:01:39) Specifically, how often do the top 100 (00:01:41) YouTubers post? Well, we individually (00:01:44) took each of the 100 channels from my (00:01:45) list, downloaded all of its uploads, and (00:01:47) then divided that by the number of (00:01:49) months that channel existed. And after (00:01:51) doing that to all the channels, we were (00:01:52) left with this graph. The vertical axis (00:01:54) shows us the number of videos posted per (00:01:56) month, the horizontal showing us the (00:01:57) years. Now, in the early days, we can (00:01:59) see today's top 100 creators didn't post (00:02:01) much, one to two videos a month. But (00:02:03) that was all until 2010, where this (00:02:05) happened. Posting volume increased (00:02:06) almost as much as Minecraft let's play (00:02:08) videos popularity. The average, which is (00:02:10) this blue line, peaks in 2015, where the (00:02:12) top YouTubers were posting an average of (00:02:14) 13 videos per month. And the median, (00:02:16) which is the red line, which is often (00:02:18) more reliable because median doesn't get (00:02:20) skewed massively by big outliers, peaked (00:02:22) in 2018 with top channels typically (00:02:24) posting about seven videos a month. But (00:02:25) what's really interesting is if we keep (00:02:27) drawing out these graphs, we can see (00:02:29) both the median and the average go down. (00:02:32) And if we just look at this graph, which (00:02:34) shows us data from the last year, we can (00:02:35) see the 100 most subscribed YouTubers (00:02:37) typically post between one to five (00:02:39) videos per month. Now, the average (00:02:41) number of uploads per month is six, (00:02:42) which is a little skewed from some (00:02:45) channels down here that post a ton of (00:02:46) videos per month, with the median, which (00:02:48) is my preferred stat to pay attention (00:02:49) to, sitting at about three. But what is (00:02:52) this actually telling us? Well, it seems (00:02:54) like the top creators definitely go for (00:02:55) a quality over quantity approach. But if (00:02:58) we go back to this year-on-year graph, (00:02:59) it looks like before switching to a less (00:03:02) frequent upload schedule, the top (00:03:03) creators posted a lot of content to find (00:03:05) their voice and refine their skills. And (00:03:07) so using this data, if we were to build (00:03:09) out a checklist that would help you (00:03:10) model the top creators, the first item (00:03:13) on the list would be quality over (00:03:15) quantity. Specifically, upload one to (00:03:17) five videos per month. But the caveat is (00:03:19) in order to get there, you might need to (00:03:21) first lean into quantity to develop your (00:03:24) skills and find your voice. But that (00:03:25) begs another question. When the top (00:03:27) YouTubers do upload, what day do they (00:03:30) upload on? To figure that out, I went to (00:03:32) the Mr. Beast channel, then used our Big (00:03:34) Brain software to download every single (00:03:35) one of his videos publish dates, then (00:03:37) converted those publish dates into days (00:03:39) of the week, and tallied them up to find (00:03:40) Mr. Beast's preferred publishing day, (00:03:42) which is Saturday, in case you're (00:03:43) wondering. And then from there, all I (00:03:45) had to do was spend the rest of my (00:03:46) weekend doing this exact same thing to (00:03:47) the other 99 channels on my list. And (00:03:49) look, after hearing this, I know what (00:03:50) you're thinking. Marcus, how are you (00:03:53) still single? (00:03:53) >> Answer: I don't know. But what I do now (00:03:56) know is the most common upload date for (00:03:58) the top YouTubers. So here's that graph, (00:04:00) which initially appears about as useful (00:04:02) as a waterproof tea bag until we get to (00:04:04) 2021, where I want to draw your (00:04:06) attention to this glass cyan line that (00:04:08) jumps out above the rest. Now that line (00:04:10) represents Saturday. And if we look at (00:04:12) data from just the last year, you can (00:04:14) see that about 26.5% of the top (00:04:16) YouTubers videos are uploaded on a (00:04:18) Saturday, which makes sense. If you want (00:04:20) viewers to binge your videos, they (00:04:21) probably have more free time on the (00:04:23) weekend compared to weekdays. And so, if (00:04:25) we were to continue building out our (00:04:26) checklist, uploading on Saturday would (00:04:28) be the next point. Now, in a second, I (00:04:30) want to show you some of the graphs and (00:04:31) analysis I did for the words top 100 (00:04:34) YouTubers actually speak in their (00:04:35) videos. But before I can do that, one (00:04:37) last question that needs answering about (00:04:39) their videos themselves is how long are (00:04:41) they? Well, using similar methodology to (00:04:43) the previous graphs, we generated this (00:04:45) one, which you would have seen from (00:04:47) earlier. TLDDR. You can see here typical (00:04:49) video links are up higher than a Snoop (00:04:50) Dogg impersonator. But if we just look (00:04:52) at this graph, which shows us the last (00:04:53) year of data, we can see top YouTubers (00:04:55) posting a range of different video (00:04:57) lengths, some over 4 hours long. But (00:04:59) those amongst you who use your eyes to (00:05:01) see things probably picked up on the (00:05:02) fact that most videos seem to be between (00:05:04) 2 and 40 minutes in length. If we dial (00:05:06) things in a bit more, we can see the (00:05:08) average video length is 34 minutes with (00:05:10) the median being 19.9 minutes. So, if (00:05:13) you want your videos to m the top 100 (00:05:15) YouTubers, I'd aim to keep your videos (00:05:16) around 20 minutes long. So, let's add (00:05:18) that to our checklist. But now we've got (00:05:20) the highle video related stats out of (00:05:22) the way. Let's analyze the top 100 (00:05:24) creators actual speech. And I want to (00:05:26) start by looking at how much the top 100 (00:05:28) YouTubers speak, aka words per minute. (00:05:30) It's a simple calculation. Basically, (00:05:32) our software looks at the transcript of (00:05:33) a video and counts the number of words (00:05:35) in it, and then it divides that number (00:05:36) of words by how many minutes long the (00:05:39) video was, and that gives us words per (00:05:41) minute. And if we now look at this graph (00:05:42) that just shows us data from the last (00:05:44) year, you can see the average words per (00:05:46) minute is 134 with the median being (00:05:48) about 156. To put that into perspective, (00:05:51) the average human speaks at about 110 to (00:05:54) 150 words per minute. But there's a big (00:05:55) butt. See, those amongst you who use (00:05:57) your brains to think things might have (00:05:59) thought, but Marcus, if we just take the (00:06:01) amount of words in a video and divide (00:06:02) that by the number of minutes, if the (00:06:04) video has a lot of silent patches of no (00:06:06) talking, then the words per minute score (00:06:07) is going to get skewed. It's going to (00:06:09) appear like the creator is talking a lot (00:06:11) slower than they actually are. And you (00:06:12) would be right. It's one of the reasons (00:06:13) I believe the average is much lower than (00:06:15) the median here. See what interests me (00:06:17) about this graph though. We can see (00:06:18) these bars down here from 158 words per (00:06:20) minute to 197 words per minute stick out (00:06:23) from the rest. And from doing some (00:06:25) manual analysis where I actually took a (00:06:27) bunch of videos, manually cut out all of (00:06:29) the silences to find the true words per (00:06:31) minute score. These numbers seem a (00:06:32) little more realistic, particularly the (00:06:34) 168 to 187 bars. And so if we were to (00:06:37) continue building out our checklist and (00:06:38) you wanted to model the top YouTubers, I (00:06:40) think I'd actually break words per (00:06:42) minute into two separate points. The (00:06:44) first point is more about how frequently (00:06:46) the top YouTubers speak in their videos. (00:06:48) Like how many actual words are in their (00:06:50) videos on average per minute, and that (00:06:51) looks to be about 156. But if we're (00:06:53) looking more to understand how fast (00:06:55) YouTubers are speaking, like the rate at (00:06:57) which words come out of their mouth when (00:06:59) they are talking, I think somewhere (00:07:00) between 168 and 187 words per minute is (00:07:04) a more accurate estimation. But that (00:07:05) brings us to another question. See, when (00:07:07) the top 100 YouTubers are speaking, how (00:07:10) complex are the words they're saying? (00:07:12) Well, we use the fleshkincade (00:07:13) readability test to measure how (00:07:15) difficult transcripts were to (00:07:16) understand. The FKGL test was created in (00:07:18) 1975 for the US military to assess the (00:07:20) complexity of their training manuals. (00:07:22) Now, the exact mathematical equation the (00:07:24) test uses is up on screen, but for all (00:07:26) the 1970s US Navy recruits among us, (00:07:28) here are some examples of it in (00:07:29) practice. This first passage of text is (00:07:31) long with complicated words, giving it a (00:07:33) score of 18.5. Compare that to this (00:07:36) sentence, which only scores 4.1, which (00:07:38) you can see is about equivalent to a (00:07:40) sixth grade reading level. Much easier. (00:07:42) Geopolitical jokes aside, though, I'll (00:07:43) put some examples up on screen so you (00:07:45) can pause the video and look at them. (00:07:46) Now, when we analyzed all the (00:07:47) transcripts from the 93,000 videos from (00:07:49) the top YouTubers, we found that for (00:07:51) most of their history, top YouTubers (00:07:52) scored between three and four. But in (00:07:55) the last year, we can see it's dipping (00:07:56) faster than my Australian Wi-Fi during a (00:07:58) Zoom meeting. And if we look at this (00:07:59) graph which shows us last year in more (00:08:01) detail, we can see the most common (00:08:02) scores are between 1 and five. And so to (00:08:05) model the top 100 creators, the next (00:08:06) point of our checklist would be keep (00:08:08) your reading level under five. Aka, the (00:08:10) average elementary school student should (00:08:12) be able to understand it. Now, I want to (00:08:14) quickly jump in here cuz there's an (00:08:15) important caveat you need to bear in (00:08:17) mind. See, just because something shows (00:08:18) up in the data I've shown you so far (00:08:20) doesn't necessarily mean that thing (00:08:23) caused the big YouTubers to be so (00:08:24) successful. It can be easy to mix up (00:08:26) correlation and causation. For example, (00:08:28) all the pro basketball players are tall, (00:08:30) but that doesn't mean basketball causes (00:08:32) tallness. But the reason I still find (00:08:34) all this data very interesting is by (00:08:36) definition, the top 100 YouTubers are (00:08:38) very successful. And we know that they (00:08:40) have spent thousands of hours and (00:08:42) millions of dollars testing every (00:08:44) possible way to maximize their content (00:08:46) and views. And when a bunch of these (00:08:47) YouTubers independently all start doing (00:08:50) the same thing, whether that's speaking (00:08:51) below a certain grade level, using (00:08:53) certain types of words, speaking at a (00:08:55) certain rate, it signals to us that (00:08:57) maybe there's something inherent about (00:08:58) doing that thing that gets more views. (00:09:01) In saying that though, we can't get (00:09:02) lazy. We still need to think about this (00:09:03) data in context and use our common (00:09:05) sense. But that led me to another (00:09:07) question. When it comes to tone, do the (00:09:09) top YouTubers primarily speak about (00:09:11) things in a positive way or a negative (00:09:13) way? Well, to calculate this, we first (00:09:15) broke tone down into buckets. We had (00:09:17) negative tone, positive tone, neutral (00:09:19) tone, mixed, and then we had no [ __ ] (00:09:20) idea. I then used my software, and by I, (00:09:22) I mean Bobby, my developer friend, to (00:09:24) run LLMs over each sentence in every (00:09:26) video in our study. For example, AI (00:09:28) scored this sentence as negative because (00:09:30) of the negative words used and the (00:09:32) phrasing of them. On the other hand, it (00:09:34) scored this sentence as positive. And (00:09:36) this one wasn't particularly positive or (00:09:38) negative, so it was scored as neutral. (00:09:40) And so then after scoring all of the (00:09:42) phrases in each of our videos (00:09:43) transcripts, our AI would then decide (00:09:45) whether that video was primarily (00:09:46) positive, negative, neutral. You get the (00:09:49) idea. And after doing that across all (00:09:50) the videos in the top 100 channels, we (00:09:52) ended up with this graph. Now, I get a (00:09:54) lot of lines here, confusion, overwhelm. (00:09:57) But the two things that I find most (00:09:58) interesting are this red line, which (00:10:00) represents negative videos, and this (00:10:02) green line, which represents positive (00:10:04) videos. And what we can see is if we (00:10:05) track positive videos over time, we go (00:10:08) from about 10% of the top 100 YouTubers (00:10:10) videos being positive all the way up to (00:10:12) around 40% with negative videos doing (00:10:14) almost the exact opposite going from 40% (00:10:16) down to around 6%. And we can see that (00:10:19) more clearly if we just look at this (00:10:20) graph which shows us videos from last (00:10:22) year. 40.7% of the top 100 YouTubers (00:10:24) tend to post positive videos, 32.9% post (00:10:28) neutral videos, 20.4% are mixed, and (00:10:30) 5.8% negative. So, if we were to add (00:10:32) this to our checklist to model the tone (00:10:34) of the top YouTubers videos, you want to (00:10:35) try to be positive number one or at (00:10:37) least neutral. But that led me to (00:10:39) another question. When the top YouTubers (00:10:41) are speaking, are they creator centric (00:10:43) or viewer centric? For example, if I say (00:10:45) you won't believe the results of this (00:10:47) YouTube study, I'm talking about you. (00:10:49) This is a viewer centric sentence. But (00:10:51) if instead I say, I analyzed all the (00:10:53) words from the top 100 YouTubers, I'm (00:10:55) now talking about the same thing, but (00:10:56) I'm talking about myself. It's me or (00:10:58) creatorcentric language. So the question (00:11:00) is how narcissistic are the top (00:11:02) YouTubers? And if we just look at the (00:11:03) last year, we can see narcissism is (00:11:05) dunking on whatever the opposite of (00:11:07) narcissism is. Is that what is that? Is (00:11:09) that selflessness? In all seriousness (00:11:11) though, 60% of the top YouTubers are (00:11:13) primarily creator centric. They use eye (00:11:15) language more in their videos with 40% (00:11:17) being viewercentric. They use you (00:11:19) language in videos. A caveat that is (00:11:21) worth bearing in mind here though, and (00:11:23) this goes back to our don't blindly take (00:11:25) all data on face value point, but the (00:11:26) majority of top YouTubers are (00:11:28) entertainment channels primarily. And so (00:11:30) I wouldn't be surprised if viewercentric (00:11:31) language was a bit more common in (00:11:33) education videos. But looking at the (00:11:34) differences between education and (00:11:36) entertainment, that's for another video. (00:11:37) And so now let's go through our full (00:11:39) checklist and add this to complete it. (00:11:41) First, when it comes to posting videos, (00:11:43) post between 1 to five videos per month. (00:11:45) And if you want a specific number, the (00:11:46) most common upload frequency is three (00:11:48) videos per month. But bear in mind, the (00:11:50) data seems to indicate the top YouTubers (00:11:52) posted a lot of content and mastered (00:11:53) their craft before slimming down their (00:11:55) upload schedules. When you do upload, (00:11:57) post on Saturdays. I'd aim to keep your (00:11:59) videos around 20 minutes long. Now, in (00:12:01) terms of the amount of words to use in (00:12:02) your videos, top YouTubers have about (00:12:04) 156 words per minute, but bear in mind (00:12:06) that stat is not accurate when it comes (00:12:08) to speed of speaking. For that, my best (00:12:10) estimation is you want to be speaking (00:12:11) about 168 to 187 words per minute. Now, (00:12:14) for word complexity, you want to score (00:12:16) under a five on the FKGL test and opt (00:12:18) for a primarily positive or if you can't (00:12:20) do that, at least neutral tone. And last (00:12:22) but not least, embrace your inner (00:12:24) narcissism and speak more about yourself (00:12:26) than your viewers. But you may have (00:12:28) noticed I didn't mention anything in (00:12:29) this video about the top YouTubers (00:12:30) titles or thumbnails, which are (00:12:32) incredibly important. And that's because (00:12:34) I actually made a whole video dedicated (00:12:35) to top YouTubers titles and thumbnails. (00:12:37) I'll link it up on screen so you can go (00:12:39) check it out. So, click that video (00:12:40) because at the end of the day, if people (00:12:41) don't click on your video, they don't (00:12:43) even get a chance to be exposed to your (00:12:45) content. Massive thank you to everyone (00:12:46) who helped me make this video. Was a (00:12:48) massive project. JV, Matt, Charice, (00:12:51) Bobby, Dave, Sashil. Wouldn't have been (00:12:53) possible without these guys. And (00:12:55) hopefully I'll see you in the tile and (00:12:56) thumbnail