1. Cracking YouTube in 2017
Matt Gielen, June 21st, 2017
Additional content from:
Mark Robertson, Melina Domingues, Tim Jablonski, Phil Starkovich and TubeBuddy
2. YOUTUBE DATA INSIGHTS
1. WHAT WE LOOKED AT AND HOW WE DID IT
2. YOUTUBE CHANGES IN 2016 & 2017
3. FREQUENCY OF POST
4. VIDEO DURATION
5. VIEW DURATION
6. VIEW VELOCITY
7. METADATA
8. SUGGESTED VIDEOS
9. FINAL THOUGHTS
4. • Researched 1096 channels.
• Looked at 36 data points per channel.
• All channels Oauth’d via TubeBuddy.
• All data delivered to Little Monster was
anonymized in advance.
VOLUME OF RESEARCH: CHANNELS
Data from YouTube analytics via TubeBuddy.
5. • Researched 69,310 videos.
• Looked at 38 data points for each video.
VOLUME OF RESEARCH: VIDEOS
Data from YouTube analytics via TubeBuddy.
6. THE LITTLE MONSTER TEAM
Thank you to the Little Monster team whose hard work and dedication to helping people build their audiences is reflected in this work.
TUBEBUDDY
Thank you to Phil Starkovich and the TubeBuddy team for the data used in this report. In total we looked at over 2.5 million data points
provided (and anonymized) by TubeBuddy.
Notes
•Median vs. Average: For certain metrics we chose to use the median value of a set instead of the average. We did this because several
sets, for example, channels with more than 1mm subscribers, had outliers that drastically altered the data.
•Groupings: For the channel data we primarily looked at the data across four groupings. This is because comparing the views on a
channel with 1,000 subscribers versus a channel with 1,000,000 subscribers presents a number of issues. For video data, we used the
average 30 day viewership as a percentage of subscribers to more accurately create an “apples to apples” comparison.
•Data: All data is from videos posted between November 1st, 2016 and April 30th, 2017. The last 6 months have seen significant
changes to the YouTube algorithm, and this data is most reflective of those changes.
•Suggested videos: For suggested video counts we used data pulled from the API. The API delivers results back as if it is a “logged out”
viewer. This is an incredibly important factor as the “logged in” view likely has many more viewer specific metrics influencing suggested
videos. In addition, we use “suggested” to mean both “Suggested” and “Recommended”.
THANKS & NOTES
8. SUGGESTED VIDEOS COLUMN CHANGES
Screen shot from Phil Defranco’s video: https://www.youtube.com/watch?v=fv5xWoyUWac
• New videos get a higher frequency of
suggested videos with the “New” tag.
• YouTube no longer reserves first 4 spots
for the uploader.
• Videos older than 1 week have fewer
“New” tags.
9. SUGGESTED VIDEOS COLUMN CHANGES
Screen shot from Coleen Vlogs’s video: https://www.youtube.com/watch?v=3IRWZ1HjC-Y
• New Suggested Column Unit “Mix” (it’s a playlist).
10. • March: Advertisers pull advertising in backlash to WSJ hit job on YouTube.
• March & April: YouTube/Google overreact, de-monetizing thousands of “controversial” videos.
• Advertisers are beginning to come back.
AD-POCALYPSE + DEMONETIZATION
Data from YouTube analytics
11. • YouTube launches YouTubeTV for $35.
• The great unbundling is well underway.
Who will win?
• Independent creators continued to be
pushed to the sidelines in quest for
“Watch Time.”
YOUTUBE IS THE NEW TV?
Screenshot from YouTube.com
13. • Channels with 200k+ subs post far more
frequently than smaller channels.
• These numbers indicate the new
‘minimum’ for success on YouTube is
now 2 - 3 videos per week.
MORE VIDEOS = MORE VIEWS.
Data from YouTube analytics via TubeBuddy.
14. • Breaking down our channel analysis by monthly views supports the notion that more videos = more views.
• Channels that are generating more than 20mm monthly views are posting videos twice as often as those in the
1mm - 2mm monthly views range, and 64% more often than the 2mm - 5mm range.
MORE VIDEOS = MORE VIEWS.
Data from YouTube analytics via TubeBuddy.
19. • Smaller channels tend to post longer videos.
• But they get shorter average view durations than larger channels.
LONGER VIDEOS DO NOT LEAD TO MORE VIEWERSHIP*
Data from YouTube analytics via TubeBuddy.
20. • Videos with average view durations between 5-8 min have the highest percentage of 1st day viewership.
• This is likely a combination of viewer choice and YouTube promotion.
VIEW VELOCITY IN RELATION TO DURATION IS A CURVE
Data from YouTube analytics via TubeBuddy.
21. • For 1, 2, 7, and 30 Day view ranges, videos with an avg. view duration of 5 - 8 min. perform best on avg.
• This suggests longer videos lead to shorter sessions and/or more exits.
• Also potentially suggests these videos get less clicks in the first few days.
BEST AVERAGE VIEW DURATION = 5 - 8 MINUTES
Data from YouTube analytics via TubeBuddy.
23. • Channels with 1mm or more
subscribers on average generate 3.5x
more viewership than channels with
200k - 1mm subs.
• Channels with 200k - 1mm
subscribers generate 3x channels
with 100 - 200k subs.
SUBSCRIBERS STILL DETERMINE VIEWERSHIP
Data from YouTube analytics via TubeBuddy.
24. • Subscribers are still the primary source of 1st day viewership.
• Larger channels get a larger portion of their audience (expressed as total views/subscribers) in the first day.
• However, on average by day 7, channels with 1mm+ subscribers begin to fade in view velocity.
LARGER CHANNELS GET THEIR AUDIENCE
Data from YouTube analytics via TubeBuddy.
25. • Channels that generate more than 2 million monthly views get a higher percentage of views in relation to
their subscriber counts when compared to smaller channels.
• These channels are generating 40% more views in relation to their subscribers than smaller channels.
VIDEOS THAT GENERATE THE MOST VIEWS OVER 6 MO PERFORM WELL EARLY
Data from YouTube analytics via TubeBuddy.
26. STRONG CORRELATION OF 7 DAY VIEWS TO 30 DAY PERFORMANCE
• At all sub ranges we see a strong correlation between 7 day viewership and 30 day performance.
• This combined with other data seems to indicate the YouTube algorithm looks at several periods to
determine whether or not to promote a video: 1 Day, 2 Days, 7 Days and 30 Days.
Data from YouTube analytics via TubeBuddy.
28. • There is no correlation between title length, description length, or the number of tags and views.
THERE ARE NO MAGIC KEYWORDS
Data from YouTube analytics via TubeBuddy.
29. LENGTH OF TITLE DOES NOT CORRELATE TO VIEWERSHIP
Data from YouTube analytics via TubeBuddy.
30. NUMBER OF TAGS DOES NOT CORRELATE TO VIEWERSHIP
Data from YouTube analytics via TubeBuddy.
31. COMMENTS DO NOT CORRELATE TO VIEWERSHIP
Data from YouTube analytics via TubeBuddy.
32. HOWEVER…
Metadata impacts RELEVANCY.
Relevancy is an extremely important factor in the YouTube algorithm.
From the Google white paper “Deep Neural Networks For YouTube Recommendations”:
“We consistently observe that
users prefer fresh content,
though not at the expense of
RELEVANCY.”
34. • Since we’re using the “logged out” view, by pulling data through the API, this makes sense.
• Appears to be largely driven by Frequency of Post & Average View Duration.
BIGGER CHANNELS ON AVERAGE GET MORE OF THEIR OWN VIDEOS IN SUGGESTED
Data from YouTube analytics via TubeBuddy.
35. • Smaller channels generate more views (percentage wise) via search.
• Larger channels generate more views (percentage wise) via suggested.
• This suggests keyword based optimization is far less important than content based optimization.
LARGE CHANNELS GET MORE VIEWS THROUGH SUGGESTED THAN THROUGH SEARCH
Data from YouTube analytics via TubeBuddy.
36. • Channels that generate more views as a % of their subscribers have more videos in suggested videos.
• This reinforces the importance of getting your own videos featured in suggested videos for the content you
create.
STRONG CORRELATION IN DAYS 1 & 2 TO NUMBER OF CREATOR VIDEOS IN SUGGESTED
Data from YouTube analytics via TubeBuddy.
37. • On average shorter descriptions appear to yield more placements in suggested.
SHORTER DESCRIPTIONS = MORE VIDEOS IN SUGGESTED?
Data from YouTube analytics via TubeBuddy.
38. • On average shorter titles appear to yield more placements in suggested.
SHORTER TITLES = MORE SUGGESTED?
Data from YouTube analytics via TubeBuddy.
39. • Channels with a large # of tags/video tend to get more of their videos featured in suggested. But…
• This correlation could be coincidental and not causation. But…
• If it is causation - it shows the importance of metadata for RELEVANCY.
NUMBER OF TAGS CORRELATE TO NUMBER OF SUGGESTED VIDEOS
Data from YouTube analytics via TubeBuddy.
40. • The strongest correlation between # of videos from the creator in suggested is Avg. View Duration.
IT ALL COMES BACK TO WATCH TIME
Data from YouTube analytics via TubeBuddy.
42. • Every video starts with the question: what should we make? The answer to this question, whether you have 1
subscriber or 50 million subscribers will be the single biggest determining factor that video’s viewership.
• This demands deep knowledge of your audience, your content and the YouTube platform.
• Thumbnails, titles and the topic of a video will determine it’s click through rate, which will determine that video’s
view velocity. Choose carefully.
• Playlists appear to be relevant again, despite a decade of user behavior that indicates little interest in watching
videos in playlists. Ignore them at your own peril for now.
• YouTube still does not provide creators with the data to serve YouTube or audiences better. Creators require
thumbnail a/b testing and data, title a/b testing and data, session duration metrics, and viewer flow models.
• Large channels are still entering rapid death spirals due to how the YouTube algorithm works.
• Changes to the algorithm in the last 6 months largely appear to be to the benefit of large channels that post
extremely frequently.
OPTIMIZATION STARTS ON THE WHITE BOARD