Waxy.org is the sandbox of Yesterday, YouTube refreshed their leaderboards and a strange new video became the Most Viewed Video of All-Time, topping the world-famous Evolution of Dance video. With 89 million views, the new winner is a fan-made music video. But did it get there legitimately?
In early 2007, a popular Italian music community called Qoob.tv announced a contest for the Brazilian band CSS, in which fans could remix green-screen footage created by the band to create their own video for their song "Alcohol." Instead, an Italian music blogger and photographer named Clarus Bartel remixed the "Alcohol" footage for a different song, Music Is My Hot Hot Sex, the song made famous by the iPod Touch ad. Update: See our interview with the video's creator.
Since he uploaded the video to YouTube, it's accumulated a staggering 89 million views, at an average 265,500 views per day. (Though, as you'll see below, most of those viewers were in the last two months.) Not only would this make his video the most-viewed of all time, defeating runner-up The Evolution of Dance by over 12 million views, but it's also added more views-per-day than any video but Britney's latest single.
The commenters on the CSS video are baffled, many of them accusing foul play. I decided to look into it, to see if there was a method to determine if the number of views were somehow false.
One method of detecting suspicious view counts on YouTube would be to compare the ratio of social activity to the view count. If the number of ratings, comments, and favorites are much lower than other videos with similar views, then it's possible that the numbers have been artificially inflated.
Using the YouTube API, I retrieved statistics for the top 500 videos. I chose to compare the number of ratings, because comments can be turned off or removed by video owners and the number of favorites was unavailable in the API. The spreadsheet is below.
Download in Excel or CSV format.
If you look at the ratio of views-to-ratings, you'll see the CSS video has a very unusual 21,487-to-1 ratio. In other words, for every 21,487 views, someone leaves one rating. To compare, the average ratio for every other video in the top 10 is a more reasonable 590-to-1. This ratio is representative of other popular videos, too; the median ratio for the entire top 500 is 545-to-1. Skimming down the list, you can see some other strange videos that might raise eyebrows.
Many of the anomalies are videos embedded on extremely high-traffic sites, resulting in huge viewership but with very little social activity. For example, this video linked directly from Google Maps or this Rihanna promo ad from from her Myspace. In these cases, you can usually click on the "Links" on the video page to see where the traffic's coming from. But for the CSS video, and several others in the list, there's no clear source of where the external traffic could be coming from.
Using numbers pulled from around the web, I've found a few historical stats for the video. In the last few weeks, the CSS video is growing by 6-10 million views every five days.
February 25: 64,557,634 *
March 1: 76,216,419 *
March 3: 83,244,120 *
March 5: 89,174,590
Is this a guarantee that the numbers were faked? No, there's still a possibility that this torrent of traffic is coming from a legitimate external source, but it seems increasingly unlikely. I'm pretty sure it's either a bug or some form of cheating, but I'd love to hear theories.
(Huge thanks to Philip Rogosky for the tip and ongoing research.)
Update: With Philip's help, we've interviewed Clarus Bartel about his clip. After hearing his side, I don't believe he's involved in faking the numbers. That said, the numbers are still very suspicious. I'm trying to get in touch with YouTube to find out more about their methodology.
March 7: YouTube removed the CSS video from the Most Viewed chart, though didn't adjust the numbers on the video itself. No word from YouTube about why, but presumably they're investigating it.
March 18: After YouTube reinstated its place on the leaderboard, the video's creator removed it from the site. More here.
4:18 AM
nice detective work. sounds pretty fishy to me.
5:12 AM
definitely cheating. I've always found the amount of cheating on youtube staggering.
from inflated view counts to letting people misrepresent a video in the video thumbnail.
while most users take youtube's metrics pretty seriously I don't think youtube care one way or the other. They still are getting page views and sites inflate their traffic all the time.
Because of the video thumbnails I've found the "most viewed today" page useless. Youtube in general seems to have little interest in filtering out the good content from the bad.
5:37 AM
There appear to be other OMG HAX videos not shown in your infodump:
http://www.youtube.com/watch?v=12Z3J1uzd0Q
61,888,445 views, 8327 ratings
6:43 AM
Also, you should have noted that it is a crappy video. There's nothing noteworthy about it to be considered viral as EOD was, just a popular song and some crappy iMovie effects. Definitely hacking.
7:28 AM
Maybe they've managed to get a hidden iframe containing the page into one or more popular sites?
XSS and malware could also be used to push a video up the rankings - [consults lawyer] - not that I'm suggesting anything so dubious is going on in this case.
11:07 AM
Im not sure on this one.
It does seem wierd, but not impossible
It was an Ipod touch advert, so it got advertisement on the TV,
plus Brazil is a large country
and because it is now number 1, people watch it, and dont watch evolution of dance because they have watched it before.
but has like 25million views in 10 days.
might just be one of those things
11:10 AM
Also it o many views because it had '??? VOTE OBAMA??' in the name
and some of the tags include, sexy, sex, barack, obama, ipod, hot, computer
which will all be very popular searches.
so i dont think it was hacked.
11:31 AM
The CSS video didn't rank well for any of those searches, however. And keep in mind the sheer scale we're talking about... The will.i.am Yes We Can video's been seen worldwide, and it still only has 5.5 million views. The CSS video received more than that in the last two days.
1:13 PM
I did see it on Reddit the other day --> Some of it's publicity is the --> LOOK - a video overtook EOD... (And it's crap).
Cheating might be a good call though.. Although as someone says - brazil is a big country -
1:41 PM
surely, just having 'hot sex' in the title will ensure that it will appear on many results, given 'normal' internet video-based content searches.
2:15 PM
If you test it yourself on YouTube or a web-wide search engine, you'll see that the CSS video doesn't appear for any sex-related queries.
6:47 PM
I can't even imagine something being viewed that many times. 89 million is a HUGE number. The title, I am sure, helps to push along the views.
5:13 PM
if you notice on your spreadsheet, the video "xxx" has about 3000 views per comment. It is possible that since the song is in the ipod touch ad, that asian viewers of the video, i.e. china, are running up the views
9:37 PM
why not just be happy for CSS? it finally kicked out emo ho's MCR, rihanna the retarted, and SOULJA BOY! (along with avril, and that dance thing) this is actually something GOOD! come on! youtube is just being a prick caus it just CAME-OUT-OF-NOWHERE success
12:21 PM
not to mention that every time a website like yours mentions it, another shitload of people go and check it out.
4:13 AM
I could believe it's a genuine view start number.
Being number one, having that title, having those tags, being promoted on Reddit and now all the controversy about it.
I've said view start number, not views, because I've not watch even 1/4 of it.
But like I clicked on it, it just will be counted as a view, probably.
[The Rojo Newsletter talked about it and I was kind of curious, so I clicked anyway.]
It would be worth while to analyse this case more closely.
How was it promoted ? some SEO lessons to be learnt !
And the second positive consequence is this question:
Any chance for getting the completedly viewed numbers on videos ?
[the number of complete views of a video are much better indicators of quality than the number of started views, if you ask me]
Pieter Jansegers
11:45 AM
i think that the comment analysis is flawed. in china most computer use is in cyber cafe, to comment on youtube you have to login, most chinese wont do this.
12:11 PM
That's entirely possible for the "xxx" video... I didn't consider that. Thanks for the feedback.
4:48 PM
The video no longer seems to be on the leader board?
2:09 PM
A agree that the most significant data point is a subjective one: this video just plain sucks.
6:54 PM
I don't want to spoil CSS fun, neither think that they have anything to do with this, but have you noticed that 16+ million views from the video came from this URL? http://hi5.qloud.com/profileBox/profileBox.swf
It seems like an application that can automatically embed videos on social networking sites... And if it can put a YouTube video on autoplay on a lot of profiles, maybe you can do the trick.
2:32 PM
Pode até ser uma fraude o número de acessos, mas a musica e o vídeo realmente são bons alem do charme e beleza das garotas.
3:09 PM
Cool research — thanks for sharing it and digging in deep.
@Andy, do you have a tutorial on how you used the YouTube API then made this spreadsheet? I didn't know it could be used to acquire stats like that.
Many thanks in advance!
5:44 PM
Remember, youtube is illegal in China, so throw the views from there right out, cyber cafe/logged in or not.
I would say this is a great example of view fraud.
i would say that we should all learn from their example.
i think we should all do the same, cause youtube to crash from global bandwidth saturation (taking a few small wired countries with it) and destabilize google's stock price, bringing about a stock market crash, financial collapse, and the ushering in of the apocalypse.
Or, maybe you'd just have the most viewed video on youtube, but....THAT is just WAY to AMBITIOUS!!
Lulz!
8:01 AM
@Torley I just used the API for searching videos to grab 50 videos at a time ordered by view count, then grabbed the date and number of views for each. Easy.