K-pop - Youtube ads spell the end of meaningful view counters

Published August 26 2019, updated September 28

For some time, Youtube has allowed normal videos, posted by any person, to be used as an advertisement on Youtube itself, enabling people and companies to boost exposition and view counts without the need for a different setup or any predefined restrictions: Ads can be as long as you want (people have reported unskipable 1 hour long ads), and about really anything, if you are willing to pay for the ad placement or even better, participate on Youtube's True View program (nearly 10x cheaper view per ad for Music Videos, check more about True View on this link).

However, either because it was not seen as useful, poor adoption or because Youtube didn't actually count those views (as an ad) to the main view counter, there was little disruption as a whole, and people were able to consider that view counters, as inaccurate as they are with people repeating or bot manipulation away, still followed a certain rule. The following, for instance, has been always the viewership trend of a normal MV for years:

Red Velvet's Zimzalabim release followed a pretty standard and normal curve

(fig 1. Red Velvet's Zimzalabim pretty standard curve)

Videos have a high peak on release and slowly fall to a certain normal (usually related to the size of the fanbase) never to peak significantly again (with small bumps when a group have a comeback).

In fact, this trend is so common, that out of the nearly 5000 videos we track, this is the average curve most MV's do during a 2 year run:

Historical average of all registered videos on Aoimirai. Green is observed current viewership, Blue average viewership, Red expected viewership given an exponential formula

(fig 2. Averages and expected behavior of viewership on 2 year interval)

But things start to get interesting already on figure 2 (You can see the full graphic with better explanation here), when observing the recent viewership trend (green): It is far above the average on the recent months, and even more than the expected behavior in red. Small peaks are to be expected, they represent anomalies introduced by all-time hits in this 2-year interval, and as of august 2019, when these graphics were generated, there were little to no super hits to answer for a nearly flat viewership for those 2 latest months in the beginning (the latest hits from Blackpink and BTS were past 4 months, and you can't even see their impact at the 4m column). So, what changed?

It is simple: something on how Youtube counts views from ads changed. It seems that more (or all) views are being counted, and thus, any agency willing to dish cash to promote/expose their artists, can seriously inflate the views on their videos. On previous studies on how bots and fans could change the view counter, we found that it were little more than 5% of all-time views, which would, at the most optimistic value, add some 10 million views to a months old video. However, with the power of ads (and money), videos are getting nearly a million views per day from what we can only call unwanted viewers.

Rolling Stone run an article on September 2019 on this practice (here), showing that to make it worst, Youtube's True View program aimed at promoting Music Videos can give you extra views almost 10 times cheaper than their normal ad policy.

Here are some examples that illustrate how serious counting ad views can be on a video viewership:

Momoland's CM for Poyo, Banana Chacha, show clearly how its addition to the ad rotation threw the view counts to the sky

(fig 3. Momoland's CM for Poyo, Banana Chacha)

This video, created by Momoland as a commercial, had little to no views (average 20k per day since release, which is already a lot considering its a repetitive children's tune). Normally, it would just continue that and be quickly forgotten to the realms of videos that can't get a hundred views a day. But about 3 months after its release, the view counter starts behaving erratically, and then stabilizes at an almost constant 800.000 views a day for around 9 weeks (Sept 28). The video, which had less than 1 million views up to then, managed to get past 40 million views. 

As a commercial video that was always commissioned to be an ad, its not surprising and would not even be a compromise to view counters, we can just ignore CMs and keep track of real music videos. The issue, however, is that agencies took note of how easy (albeit somewhat expensive) it was to promote a video/music with Youtube ads, so quickly after this, other agencies started using it. Here is a good example:

Everglow's Adios and its anomalous ascencion

(fig 4. Everglow's Adios - An anomalous ascension. By september 1st the video would drop to less than 500,000 view/day suddenly)

At first look, it doesn't look that anomalous: it started high, then leveled off low. It looks like it, but it isn't. First, it took it too long to reach peak view, which is not normal for a non-debut video (Everglow at this time already had some exposition thanks to its noteworthy debut). What causes the illusion, on purpose or not, is the fact the video entered the ad rotation one or two days after release, so the peak from its release intertwined with the peak from the ads. However, what really tells there is something fishy going on, is the fact that this music, that didn't enter a single music chart top 10 ever since release, maintains its place in the top 3 of the most watched video. To make this even more clear, it was released on the same day as Umpah Umpa from Red Velvet. No matter how one could argue this or that song is better, all songs always have the same trend as seen on figure 1 - including Umpah Umpah. To further add facts to it, an alternate version and a dance version of the same song were released shortly after, and they do not exhibit this behavior, instead, showing the normal peak-then-level trend, and low viewership compared to this main video. The same music, with trends on different videos contradicting, means one thing: the main video is getting an external influence: ad promotion. Again, keep in mind this song didn't get into any top 10 chart (Umpah Umpah, which got less than half the views, did).

But one could still argue that the music was a surprise and people started noting it with a few days delay (which is not the case, otherwise the alternate and dance versions would also show it), so there must be other signs that a song is being promoted (and the unsolicited views counted) that are not just the strange view/day trend. And there is: View/Like behavior.

Ads do not allow you to like or dislike (unless you really want to get into the video just to dislike it, like what happened with Banana Chacha, breaking the record for a K-Pop song like/dislike ratio all thanks to this), so all those views added by ad promotion don't aggregate likes. Enters an analysis of View/Likes for the latest videos (with big view/like) as of August 25:

Top anomalous views as of early september 2019

(fig 5. Top anomalous View/Like in the last 3 months - yellow are guaranteed to be being used as ads due to off-normal view/like)

Take a good look at Nature's "I'm so pretty" (fig 6). There is not even a need to show the graphic of view/day when the View/Like is that off trend (1080 view/like), but it looks very similar to Banana Chacha, except it lasted two weeks. It is also extremely "suspicious" that is stopped exactly at 10 million views and stopped being viewed. You will notice that the View/Like distribution is completely anomalous in all these videos, but some are not marked as being used as ads, except it's not hard to spot the ones which are: Total views. Videos with view counters lower than about a couple million certainly did not get the boost from ad promotion. Another important factor is that the older a video gets, the higher the View/Like (old videos get less likes because everyone who would like, already liked, and the new views are repeats, thus a 5 year old video can easily have some 500 View/Like and not be suspect of being used as a ad promotion, but a new video which did not have time to stop getting likes from new valid views, should not get this anomaly).

Nature's I'm so Pretty show what happens when you dump all your money on youtube ads. All views are in the ad period, then it drops to less than 2.000 view/day. This is what 1080 View/Like looks like

(fig 6. Extremelly anomalous distribution for Nature's "I'm so pretty" with 99% of the views calculated as ad views at 1080 view/like)

You might think these View/Likes "don't look that bad", but they do. Most videos under a month hardly get to 30 View/Likes, and in fact, during the first week they often manage to stay near 10 View/Likes. While not in the top 15 at figure 5, Adios have a 42 View/Like at the moment, and it is only 8 days old. For comparison, Umpah Umpah have 11 View/Likes, which is normal. You can see how the most recent videos are doing on View/Likes at this link.

Aoimirai's ad promotion detector will flag any video that have a View/Like exceedingly above average during the first month, at which point it starts to get hard to tell if the video is just being too repeated or are back to promotion. Another easy markers for a video being promoted are view counts in excess of a couple million views simultaneously to the View/Like spike, and a spike on viewership that, as of this month, averages 800 million views / day (although different ad plans probably allow you to adjust that).

An interesting aspect of monitoring the View/Like and cross-referencing to the average that normal videos of the same age get, is that you can even estimate how many views that video get from ad promotion. Take "I'm so pretty" from Nature and its 1082 View/Like figure for a one month video. Videos in that range should be sitting in a comfortable 20~30 View/Likes. This means it have about 40 times more views than it should: divide the total views by it and you have the non-ad promoted views of 255.150 views. Based on alternate versions and their previous video, that is actually very accurate, and again point to the estimate that they paid for 10 million ads.

X1's debut video flash clearly being promoted for 14 days

(fig 7. Clear view shut down after 14 days running ads on X1's Flash video, notice it drops from 2,000,000 vpd to 400,000 vpd in two days)

Music Video counters don't mean anything taken alone, and if the agency didn't pay for promotion, it would actually mean they could have profited from ads shown before the music video; but considering Music Videos have always been about promotion and exposure, it does make sense to cash in and make it a proper ad. It is just sad that some new groups that are struggling to get know think (or rather, their agency thinks) that forcing thousands of Youtube users to sit through an unknown and unwanted music video is a positive promotion. Just a look at how many dislikes Momoland's Banana chacha started raking after it went for ad promotion, and you can tell it is not a good idea. Its doubtful that those 10 million ads for Nature converted too many people, but one can easily see some people getting mad at having to sit through a 4 minute music they didn't ask for and end up associating it with a bad vibe. Music Videos are good promotion, but not when they are forced into you.

Conclusion, if this trend is here to stay, and probably is, because there are plenty of agencies that are really bad at understanding how to promote on the internet or use Youtube (the big mature companies are well aware that using it as a free promotion that even gives you some profit from ads played on it is far better than annoying unsuspecting netizens while losing money), so we might want to take big viewership with an extra grain of salt, because making music videos in official ads just threw all the speculation on how much fans repeating a video all day, or getting some bots to inflate the views, under the bus, and those who like to compare most viewed videos are in for some extra level of confusion on fairness, since so far, the math to calculate how many views the ad promotion got is not safe, and all we can do is know it happened.

Below, a simple 8 week table of expected View/Like for a normal non promoted video, it doesn't matter if it goes viral or not. What can usually increase view/like is smaller artist with a small fan-base that causes view/like to increase even though the viewership is under a million, that happens because only real fans are streaming (multiple times a day to watch, not to manipulate) and, therefore, increase the view/like a little, but usually still within some margin. 

Weeks since launch Average expected View/Like (half of max)
1 week 5~11 vpl
2 weeks 12~23 vpl
3 weeks 18~30 vpl
4 weeks 25~40 vpl
5 weeks 30~50 vpl
6 weeks 35~55 vpl
7 weeks 40~60 vpl
8 weeks 45~65 vpl

 

A simplified formula is, starting on the 5th day: MAXIMUM expected view/like = 15 + ( factor * days-since-launch) and slowly gets factor from 2 to around in a year 0.5. A year old video should max around 250 view/like if very successful, or less. 

Full list of updates on this article, since release in August 26, 2019

Update 1: I would like to point out that all artists and videos mentioned in this article are just used as examples, not to belittle or tarnish them, specially because the point of the article is arguing about Youtube profit machine and how it is now having clear consequences on views (which are used on Music Shows to pick winners). The choice for Zimzalabim first is because it is a good recent example of a pretty normal (almost boring) viewership trend for a big group (one could see the same with a small group, but the graphic would be less accentuated and thus might fail to show the trend). The choice of Umpah Umpah and Adios later is solely because they were released almost at the same time, are both videos expected to have good runs (and they had), but show totally different trends and data; finally, the fact Nature's I'm so Pretty graphic, as well others, are not shown is because it was not the intention to flood this with a lot of graphics and you can see them (at least while they are less than 90 days old) on the site, and are just a repeat of the shown graphics. The conclusion on View/Likes is a well know indicator of manipulation of views, but is used to show how Youtube ads managed to create even bigger anomalies in an already simple way to detect view manipulation. Also, keep in mind, as mentioned at the end of the article, that View/Likes is expected to increase with time.

Updated 2: The update above just made me aware that view/day data is not preserved forever and thus Nature's I'm so Pretty data will be lost, so I decided to go ahead and add the graphic where its mentioned bellow

Updated 3: Added X1's debut video Flash to show how you can clearly see promotions. Added small table of normal expected view-per-like

Updated 4: Added link to Rolling Stone article on the subject, which were published almost a week after this article. Added list of known instances of MVs used as ad (another page). Also, check this article about True View