TubeMogul, announced the results of research today that clarifies how each of the online video sharing platforms supported its distribution platform count online video views. As the study reports, there are various methods that each site uses; some more stringent than others with views counted only as one-view/IP address, others merely count all views. It appears that from the changes that have occurred since their last release of this report, a standard for this analytical measurement has started to emerge.
TubeMogul tested various actions on 14 different video sites, including hitting refresh on a video after it starts playing, watching more than half of a video, watching a video to completion and playing embedded videos. IP addresses and cache settings were varied when necessary.
TubeMogul found that once a video starts playing, all but three sites tested logged a view no matter how much of the video is watched, reflecting an increased standardization since the study was initially published last June. Interestingly, both Youtube and Yahoo! Video, joining the majority of video sites, lessened their standards since last June, previously having IP address-based constraints. Now, both sites count everything once the video starts playing. It appears that Blip and Metacafe are lone holdouts to a stricter, IP address-based standard. There is also some spotty disagreement (DailyMotion, Howcast) on whether to count embedded views and how they are counted.
"That a standard is starting to emerge, with big players like YouTube embracing it, is great news for content creators trying to monetize their videos and relying on views data to sell ads," commented Brett Wilson, Co-Founder and CEO of TubeMogul. "That said, further transparency and standardization would be ideal.”
Here is the full report - "View" testing for various online video sites
Monday 19th of May 2008 12:00:00 AM - This study was initially released last June to great reception from both media and content creators. Here, we update the study using the same methodology (and a few sites added) to reflect current conditions, noting changes since last time in the Conclusion section.
I. Research Abstract
This report discusses the findings of an experiment designed to test which actions result in a "view" as measured on 14 online video websites. The impetus for this research was to identify the differences in view counting among popular video sites, particularly as the online video industry has not yet adopted a standard definition for a view. The implications are relevant for video advertisers, content publishers, and those that might seek to artificially inflate the popularity of a given video.
The methodology employed in this study is a starting point and by no means an exhaustive list of testable scenarios.
Four scenarios were tested in this experiment:
- Watching a video start-to-finish multiple times
- Watching a video and stopping more than half-way (but not to completion)
- Refreshing the browser after video play has begun
- Watching a video multiple times through a player embedded on another site
All tests were run from a static IP address on a single computer between 4/15/08 and 5/14/08 logged into an account different from the video author's for each site. For stricter sites, we also tested resetting the cache or being logged out. Where available, "Private" settings were utilized to prevent others from viewing the test video. The text for the title, meta-tags and description of the test video were all jibberish (reading: "prozhectblinque”) to prevent anyone from stumbling onto the video, although this turned out not to be a problem in most cases.
III. Summary of Findings
|Site||Full View||>1/2 View||Refresh (Player Starts)||Embed (>1/2 view)|
|Blip||One/IP addr.||One/IP addr.||One/IP addr.||One/IP addr.|
|Dailymotion||Count||Count||No Count||No Count|
|Metacafe||One/IP addr.||One/IP addr.||One/IP addr.||One/IP addr.|
AOL Uncut - Video was marked "Private" to prevent others from finding it via search or by browsing the site. All tested categories resulted in views counted, including embedded views.
Blip - Blip only logs one view per IP address per session, regardless of whether the cache is cleared after each play or the user is logged in or out of the site. Given that constraint, however, once the video starts playing, a view is logged whether the video is viewed in part or in full. The video used in testing was public (Blip charges for a "Private" setting), but it is clear from the data that no one else viewed the video. Initially, we thought that Blip was not counting embedded video views, but upon review this is incorrect. As in all settings, Blip counts one view per IP address per session.
Crackle - Basically, once the player starts, Crackle counts a view. Crackle logged "views" for all categories tested, including refresh, >1/2 view, full view and an embedded view. Crackle does not offer a "private" setting, but it can be inferred from the data that no one outside of our testing group viewed the video.
Dailymotion - Video was marked "Private" to prevent others from viewing it via search or by category. No views were logged from refreshing the browser after play started or from embedding the video on another site. DailyMotion does not record views that result from embedded videos.
Howcast - Video was marked "Private" to prevent others from finding it via search or by browsing the site. Refresh, >1/2 and full views were all counted. Howcast does not count views from embedded videos.
Metacafe - Metacafe informed us that they count a view if 50% of the video was viewed greater than 30 seconds, whichever happens first. However, from our testing, it appears that even multiple full-length views are not counted in full. Communications with admins at the site revealed that views are tracked by IP address, so perhaps even full views in multiple are not actually counted from a single computer. MetaCafe counts a single view that originates from embedded videos. It appears that if multiple views come from the same IP address, just the first is recorded. Metacafe does not provide a private setting, making testing difficult. We discovered that most new videos are featured on a page driven by recency, so we waited until after an initial spike in views the first day to start testing.
Myspace - Video was marked as unlisted to prevent others from viewing it through a search on the website. All tested categories resulted in views counted. Refreshes may have counted because it took approximately half of the video for the page to reload. Myspace counts all views that originate from embedded videos.
Revver - All tested categories resulted in views counted. However, it is important to note that Revver switched their methodology for counting a view in March, whereas previously they had been one of the more stringent in counting views. At that time, the site switched from a view being a completed play to being an initiated play. Revver still captures "completions" in the API data, though completions aren't published publicly. They are the only site in the study that has completions in their API, and this information reflects that the advertisement was viewed. Revver counts views that result from embedded videos.
Sclipo - All tested categories resulted in views counted, including embedded views. Sclipo does not offer a "Private" setting, so this video was live during testing.
Stupidvideos - All tested categories resulted in views counted, including embedded views. Stupidvideos does not offer a "Private" setting, so this video was live during testing. However, there was little to no outside influence on the view count.
Veoh - All tested categories resulted in views counted, including embedded views. Veoh does not offer a "Private" setting, so this video was live during testing. The video was not added to the profile's favorites page, however, and there was little to no outside influence on the view count.
Viddler - Video was marked "Private" to prevent others from finding it via search or by browsing the site. All tested categories resulted in views counted, including embedded views.
Yahoo! Video - Yahoo! does not provide a "Private" setting for videos, so there was no control over views counted for the test. However, there was little to no outside influence on the view count. Despite their former restrictive standards (Yahoo! as of last June did not count embedded or partial views), all categories tested resulted in views counted.
Youtube - Video was marked "Private" to prevent others from viewing it through a search on the website, except during testing. Despite formerly strict standards, all categories tested resulted in a view. We also attempted to "max out" the number of views, with the idea that YouTube might count up to a certain number from the same IP address (i.e. count views up to 100). We executed 130 refreshes (as defined above), all of which were counted.
Most Stringent Sites for Counting Views
|AOL Video||Less Stringent|
|AOL Uncut||Less Stringent|
|Yahoo! Video||Less Stringent|
Conclusions: A Standard Begins To Emerge
Interestingly, both Youtube and Yahoo! lessened their standards since last June, when this study was first published. Previously, Youtube only logged one view per IP address unless the video was viewed in full. Youtube also formerly counted only one maximum view for an embedded video, no matter how many times it was played from a single IP address. Now, Youtube counts everything once the video starts playing. Yahoo!'s recently re-launched video site reflects similar changes.
It appears that Blip and Metacafe are lone holdouts to a stricter, IP address-based standard. There is also some spotty disagreement (DailyMotion, Howcast) on whether to count embedded views and how they are counted.
That said, the lack of standardization presents complexity to content producers and advertisers in understanding the relative popularity of videos across video sites. To fully realize the potential of advertising models in the online video medium, increased standardization and transparency is required.