In this post, I look into how I spend my time on YouTube and how that has changed over the years. Then I wrap it up with a neat little bow by interpreting these insights against my own personal experience as an audience member of the YouTube platform.

They say that friends come and go, but the internet is for life! 
 

        I still remember the first time I ever clicked on the AOL browser to join the final frontier: cyber-space. I was half-expecting to jump right into the computer screen and catch some gnarly email waves like when Timmy Turner search and destroyed his loveletter to Trixie. Instead, I called my dad over to help me figure out how to visit a website. This wasn’t because I couldn’t find or understand the URL bar, I knew that the URL was where the magic happened. Instead, it was but because I couldn’t read. My dad’s response was “You probably shouldn’t be on the internet if you can’t read”. Fair enough dad. Or as I would have said at the time: knasdbfoiuh hhn quewioh wnqpbieguhqpie.

 

 

Figure 1: Timmy Turner barreling through cyberspace

        Cutting to a few years after I learned how to read (hello!) and a lot of my time was spent in front of a screen. Like, a lot. If you’re around the same age as me you probably have a very similar story. In fact, I bet you’re reading this on a screen right now. 
       

        What you may not have known was that some websites you were using were with you every step of the way. Recently, Google has made it easier than ever to access the data they collected on you. By visiting what Google calls your “takeout”, you can download the gigabytes of data that Google has collected on you over the years and go start mining! This includes many products available on the Google platform; including gmail, Drive, maps, and you guessed it, your YouTube data.

Soul-Searching  or Soul-Querying?

        Google officially acquired YouTube on October 9, 2006 and YouTube officially acquired me on May 4, 2009. My first search on record was for “Captain Sparklez” on October 22, 2011. Since then, I have made over 8,000 searches for everything under the sun. Some of my favorites from each of the past 10 years are listed below: 

•    2011-10-25T02:37:52.400 – “optical illusions that make you feel underwater”
•    2012-12-04T05:06:52.767  - “nostradamus predictions about gangnam style”
•    2013-01-13T23:10:15.228 - “15 prison terms you must know”
•    2014-12-01T07:25:29.533  - “Battle without honor or humanity”
•    2015-06-26T17:34:00.911  - “why do men like blobs think tank”
•    2016-08-06T00:13:51.437  - “cartoon cats marching band”
•    2017-04-19T04:58:02.741  - “how advanced ai can destroy the world”
•    2018-02-06T01:33:12.465  - “soviet national anthem bass boosted”
•    2019-05-26T19:35:18.141  - “how to teach your dog to play nice with other dogs”
•    2020-11-08T05:12:45.497  - “spongebob metal”
•    2021-02-27T15:47:18.128 – “ww2 ambience 10 hours”
 

One of the first questions I wanted to answer was how my use of the search bar has changed over the years.

 

 

Figure 2: My number of recorded searches on YouTube for a given year

        According to Figure 2, it looks like my number of recorded searches has been on the decline for some years now. While we should keep in mind that the number of searches in 2021 includes only about 3 months of data, it is still significantly fewer searches than the previous year. When I did use it, what did I end up searching for?

 

 

Figure 3: Search terms by number of times searched

        Figure 3 is funny to me for a number of reasons. As we saw in the previous plot, the majority of my searches were conducted before 2016. At that time, I loved watching videogame let’s plays which is why it makes sense that the searches in blue all represent gaming channels. The odd one out is the query for a guitar tuner, probably because I can never be bothered to buy my own.
        The vast majority of my searches (82 %) are searched just one time. After that, the number of times a unique term is searched drops off dramatically (as seen below in Figure 4):

 

 

Figure 4: Number of times I search for a unique term

 

Who Gets the Click?
 

        So far we’ve only looked into search history, but a search is only as good as the video it gets you to watch. So how many videos do I normally watch, how much time do I spend on them, and how has that changed over the years?
It looks like at some point in 2015 I deleted my previous YouTube watch history. Wojcicki only knows what video prompted me to do it, but my watch history up until that point will forever remain a mystery. Additionally, 632 of those videos are no longer publicly accessible. Therefore, I will try to make the most of the remaining ~23,000 videos to analyze. 
        Unfortunately the “takeout” data provided by Google only tells us which videos we watched and does not include personal data about my watch time. Thankfully we can still get additional data about these videos using the YouTube Data API that will help us estimate the time spent watching. A useful tutorial on how to get started with this API is given here. A no-bs starter code can be used below:

from googleapiclient.discovery import build
youtube = build('youtube', 'v3', developerKey='YOUR KEY HERE’)

# Send the request to the YouTube API. I found that I can submit
# up to 50 videos per request. A video’s ID can be found from your Google
# takeout data OR from the URL. Ex. The ID is “coZbOM6E47I” for
# https://www.youtube.com/watch?v=coZbOM6E47I
request = youtube.videos().list(part='contentDetails',id=[‘ 8d_202l55LU’, ‘k-gYpWMZm5Y’, … ‘OTHER VIDEO IDs’])
video_info = request.execute()

        Also be completely honest with you, a lot of the content I watched over the years I now consider pretty embarrassing. There could be a really fascinating blog post written about the types of videos I watched and how they changed over time, but for full transparency I’m going to keep these details out of my analysis for now. Go find your own cringe!

With all that out of the way, let’s take a high-level overview of my YouTube watching habits: 

•    On average I watch 12.5 videos per day. On median, I watch 8 videos per day. Since the average is 1.5x more than the median, this tells me that on the days I watch more videos than the median I tend to watch a lot more!
•    The highest number of videos I watched in one day was 113 on April 16th, 2020. This was on the day of an open-resource college exam that took me almost 8 hours to complete. YouTube more than likely saved me here!
•    My longest video watching streak is 49 days, and my longest streak off YouTube is 15 days when I was in another country.
•    There were 1,903 days between the start of my YouTube history and the start of this analysis. I watched at least one YouTube video for 1,605 of those 1,903 days. Therefore on any given day there was an 84% chance I was on YouTube. Put another way, I was likely to be on YouTube 6/7 days of the week.

 

 

Figure 5: Six out of seven days spent on YouTube

Just like the number of searches, the number of videos watched also seems to have taken a dive:

 

 

Figure 6: Number of videos watched over time

        By the time we break the graphs down by weeks the data starts to get pretty noisy. Therefore, we will analyze yearly and monthly data. From these two graphs, we can see that the number of videos watched (at least on this account) has slowly decreased over time. While this might suggest I’m watching less YouTube, it might also suggest that I have been reducing the amount of time on this account. I remember a few months ago I disabled the YouTube app on my phone. I also use YouTube more frequently while signed out completely. This makes it much easier to avoid wasting time on YouTube (and reduces the amount of data they keep on me).

        Looking at this data might suggest that I am watching less YouTube, but inspection of the average video’s average duration tells another story.

 

 

Figure 7: Average duration of videos watched by year

        Notice that climb in average watch time / video duration after 2017? I looked into changes in the YouTube platform around 2018 and found this YouTube news blog post that discusses how the platform’s partnership structure changed. Specifically, YouTube’s Chief Production Officer Neal Mohan and Chief Business Officer Robert Kyncl wrote:

“Starting today [January 16, 2018] we’re changing the eligibility requirement for monetization to 4,000 hours of watchtime within the past 12 months and 1,000 subscribers.”

        This was a significant difference in the way YouTube selected its partners, especially since the previous requirement was just 10,000 lifetime views. This new requirement seemed to incentivize creators to produce longer videos more frequently to maintain their partnership. As a result I, the audience, started watching longer videos.

Digital Hygene: Watching Habits and Times

        At what time of the day do I find myself watching the most YouTube? When I first grouped the number of videos by hour watched it appeared that I was watching most of my YouTube towards the end of the day (and especially late into the night).

 

 

Figure 8: Number of videos started by hour of the day

        Notice that there is a sharp decline that starts at 4 am (probably my threshold for sleep, even on a late night) and a minimum that occurs around 9 am when I watch the fewest number of videos. Now, compare this to the video’s average duration when started on a given hour:

 

 

Figure 9: Average video duration by hour of the day the video was started

        The fewest number of videos are watched at 9 am because that’s when I watch the longest videos! I tried plotting these two datasets on the same graph but because of scale differences the result was not quite what I was looking for…

 

 

Figure 10: Comparing number of videos watched by average video duration

        To help solve this visualization issue I normalized both datasets. Normalization squishes all of the data between 0 and 1, where 0 is the minimum of the dataset and 1 is the max. Therefore, I could keep the general shape of both lines without having to worry about the fact that the scale of one was minutes watched and the other was number of videos.

 

 

Figure 11: Normalized comparison between number of videos watched and their average duration

        The danger-zone for me seems to be around 3 pm (x = 15), where I start watching more long-form content until midnight (x = 0). After midnight, I generally start to watch a lot of shorter videos (especially around 4 am), when I seem to be at the end of my watching session.
 

An Addict Yes, but High-Functioning? 

        You may recall from a previous blog post that I have been closely tracking the time I spend each day learning data science skills. I was interested to see how my YouTube watch time compared to my most well-documented recent time commitment and was actually surprised by the results. Below we have the two commitments plotted against one another.

 

 

Figure 12: Time dedicated to YouTube compared against time spent on ultralearning project

This plot is very noisy and difficult to interpret, so let’s check and see how things look on a month-by-month basis.

 

 

Figure 13: Time dedicated to YouTube compared against time spent on ultralearning project 

on a monthly basis

        Much cleaner. With the exception of December, it looks like my time has been spent elsewhere on something other than YouTube and Data Science Learning since December (likely my new job and moving across states). The Pearson correlation between these variables is incredibly low at 0.137, More importantly, this tells me that YouTube is not getting between me and my Data Science Journey. Hurray!

 

Mindful Media: The Rise of the All-Knowing YouTube Algorithm        

 

Figure 14: YouTube algorithm strength over time

        The final figure in this analysis is a simple but eye-opening one for me. Depicted above is the "Algorithm Strength" which is the number of videos I watch divided by the amount of searches made over the past few years. There was a slow and steady rise from 2016 to 2020, then a sharp jump from 2020 to 2020. You probably saw this if you have eyes, but why does any of that matter to me?

        One of the things I’ve been learning more about lately is mindfulness, or the idea of being aware of what you are doing while you are doing it. If there is one thing I have not been mindful about throughout 2021, it has been my watch time on platforms like YouTube. If I was more mindful about how I use the YouTube platform and really wanted to use the platform like a tool I would probably use it like this:

1.    I have an idea of what / who I want to watch and use the search bar to find that content
2.    I click on the video I searched for and watch it
3.    I search for the next term that I want to watch OR leave the platform

        This would dramatically reduce the videos watched / searches made metric (and likely result in much less time spent on the YouTube platform). Instead what I’ve been doing has been more like:

1.    Get bored
2.    Go on YouTube and let the algorithm decide what to show me

        Most days in this past year I would just open the YouTube home page and pick what I want to watch from there. In fact, I would often open multiple videos from the home page in different tabs and watch them separately. Taking a higher-level look at this data has clearly indicated to me that a change needs to happen in the way I use YouTube. So what am I going to do about this one?

More You Less Tube

        How can I use this platform in a healthier way and regain control over my habits around it? First I have to recognize that this YouTube algorithm is a beast, and the more I feed it the stronger it becomes. Therefore, actions like deleting the YouTube app off my phone (or “disabling” it because Google doesn’t like people leaving their apps) and signing out of my account prevent YouTube from knowing what I want to see and using that to keep me on the screen.
 

Going Forward with YouTube

        Some of the best content I have ever seen has been published on YouTube. There are so many talented creators on that platform who are criminally underrated, and I am so thankful that they shared their art, thoughts, ideas, and moments with me and the rest of the world.
        Unfortunately, YouTube’s policy shift in 2018 began to incentivize creators to churn out more long-form content more frequently. As a result, I as an audience member slowly acclimated to a longer watch time per video. My lack of oversight into my own use of this platform has led me to... My goal going forward is to use YouTube as a tool by searching for videos I would like to see then watching them. It will take time to reduce my use 
        Looking back on these searches is like looking back on a digital journal that only I and everyone at Google can access. For better or for worse this platform has influenced how I treat other people, the way I think about myself, and what I believe about the world. Seeing what I was searching for, watching, and posting reflects pretty closely with what I remember about the way my life was going at the recorded times.