Independent Films by the Numbers

The marketing of Independent Films

Archive for the 'YouTube' Category

The Best YouTube Video Length? — 8 Min.

I am programming a more intense study as I write, but I wanted to close out my initial study on the best length for a YouTube video. Common knowledge has it that since most videos on YouTube are less than 3 minutes that it is best to follow the pack and make many short videos. My study of the differences between randomly selected videos and those deemed popular/most viewed has questioned this pack logic.

What I have found is that videos under two minutes are less like to be widely watched that expected by their frequency in the population, while videos 5 minutes are over are much more likely to become popular that expected by chance and their population frequency. The strongest performance beyond chance comes from 8 minute videos, which have a stunning chi-squared value of 188 on 1 degree of freedom. Scores higher than 10.8 are able to reject a null hypothesis of the observed being the same as the expected with a confidence level of 99.9%!

I have charted this patterning by plotting the runtimes of YouTube videos against the chi-squared values looking at the expected number of popular videos based upon the population frequencies for those same runtimes. To make the chart easier to understand, I made the chi-squared values negative for cases where the observed was less than expected. Enjoy.


Comments are off for this post

Research Summary on YouTube Optimization

I have been doing more research on the state of affairs with Optimization of YouTube videos. At least publicly, there appears to be not much substance published on the topic beyond common sense and the application of SEO rules to the platform.

The most notable source I have found so far are Jonathan Mendez’s blog “Optimize and Prophesize” and a post at These two site focus on tags as the means to drive viewership much in the same way that webpages are optimized for organic search.



The upshot of these blogs is that optimization should focus on tags by:

  1. Matching tag content to titles and description
  2. Using unique tags for each video (along with unique descriptions and title)
  3. Make tags relevant
  4. Use adjective to help better engage your audience
  5. Avoid use of standard stopwords using in Natural Language Processing
  6. Use as many tags as possible (although this point by Mendez’s blog was debated in the comment responses to this point, based upon a searching index’s tendency to become unfocused when too many options are given).

What you should notice in this work is twofold – First it does not transform YouTube optimization beyond standard natural SEO and it seems very anecdotal based. The method and data backing these recommendations are unproved by information to back their assertions. Second, it is unclear how to apply this advice for impact since it is unclear beyond some loose description what make a good tag beyond it should not be from an undefined list of stop words.

Beyond these tag based insights, I found these others pearls of YouTube optimization from other sources (unvalidated by hard data as well):

  1. New video do not get search preference at YouTube over older content.
  2. Many views are driven by becoming a featured video, but this process in an internal one.
  3. Inbound links are important from trusted domains.
  4. Use the social aspects of YouTube and the 2.0 web to drive viewship/li>
  5. Sex and humor sells. It is a good strategy to find strongly performing content and make a lampoon of it or create a response to it.
  6. Carefully consider the category to be used for the video to make sure it matches the content
  7. Some blogs advice that voting and comments matters, while others think this is not the case.
  8. Make you title catchy, but don’t give too much away

All in all there is not much good content on how best to market your video on YouTube beyond just making it findable by standard search. It seems to me that insights from work I have done for optimizing films for festivals, advertising messaging, and e-mail for delivery would provide some much needed heft to the YouTube optimization toolkit.

I need to plan my big test….

Comments are off for this post

YouTube Title Optimization Probed

My initial work on analyzing factors for success on YouTube has continued with a quick analysis of Titles. It is clear to me that a bigger, longitudial study is now needed with a much larger sample size. The initial work has oriented me well, but it has its limits especially since the life cycle of typical YouTube video remains in the shadows.

The initial title work has focused upon title length in characters. It is not earthshaking, but it does provide some an initial view into the psychographics of the video posters. Again, I have compared popular/most viewed distributions against a random control.

What I have found is that there is a natural mode at a length of 32 characters for the title in both segments. The random segment has a second mode with fewer characters (<25 characters), which suggests that these might be largely hastily posted video with little effort crafting an involved title. There is one last mode at 70 characters which is the saturation point of naming -- in other words the point at which names cannot be longer. Are these extra-long names attempts at making video more findable by organic search, or just the verbose efforts of passionate uploader? I don't know yet. Anyway, here is the chart... Bar chart of title lenghts by segment

Comments are off for this post

Flaws with other YouTube Optimization Recommendations

I have been researching the state of the art for YouTube optimization and I must admit it is weak.

What bugs me most with the previous work on this subject is that it lack any baseline to compare. Just because most popular videos are shorter than 5 minutes does not mean that it is wise to make your hopefully popular video shorter than 5 minutes.

Given that all videos on YouTube tend to be short, the question should be are five or less minute videos represented in the popular category more than expected by random chance sampling from the greater population. They are not. My early work suggest that it would be foolish to cut your video short to optimize its chances for popularity.

The other recommendation for optimization I found are equally poor and under theorized… the influence of tags needs to be better addressed beyond rolling over now dated search optimization strategies for tags.

Time of day and day of the week optimizations exist, but they lack rigor of quantifying why this is important and how this effects the success of videos.

All in all, we can do better.

Comments are off for this post

Number of Tags per Video on YouTube

I have often wondered about tag in video site. Do they make a difference, since they allow videos to be found? Or is the number of videos so vast that they become lost in a sea of other tags? I don’t know the answer yet, but I do know this…. there is a difference between popular/most viewed videos and my random control selection. Popular videos average more tags spread out over a wide somewhat normal distribution, while my control sample has inverse decline pattern with most videos having one or just a few.

Are the number of tags cause for improved performance or are the best performers more carefully crafted? I need to brood on how to show which is the best explanation of what I am seeing.

Number of Tags per Video on YouTube

Number of Tags per Video on YouTube

Comments are off for this post

Pattern of Run Times for YouTube Videos

Given that there are better movie lengths for festival films, I have begun to wonder if the same is true for YouTube videos.  I am not the biggest fan of YouTube, but given its importance as a grassroots distribution channel I thought I should get over my personal biases and take a deeper dive into what makes it tick.

My initial study has involved capturing data on 2,600 video submitted yesterday.  I drew the sample from YouTube’s top 13 channels (minus Movies and Music, which seem to play by different rules) for a collection of 200 videos for each channel.  Those 200 hundred videos were split evenly into two segments: popular videos for that day and a control group of randomly selected videos.

I am still early in my analysis, but it is clear that these two samples are distinctly different.   Randomly selected videos tend to skew shorter with the largest mode occurring with videos less than one minute long and all others being 10 minutes or less in length.  If you look at the distribution of this random control segment it is a classic inverse power drop off with the count of videos dropping as the lengths go up.  The Popular/Most Viewed segment has a very different form.  There is a mode at less than one minute, but it is much less pronounced than the control segment.  There is a second mode for the popular segment at 10 minutes with a tail of lengths extending out past 10 minutes.

It seems that popular videos tend to be longer than the average video, which is a pattern not unlike what I observed for short films in festivals.  Despite not be normal in distribution, the averages run times are interesting with the control group having an average run time at 2 minutes 37 seconds, while the popular segment has an average run time at 4 minutes 52 seconds (almost twice the length) .

Runtimes for a sample of 2,600 YouTube Videos

Runtimes for a sample of 2,600 YouTube Videos

Comments are off for this post