Creators Say They’ve Cracked YouTube’s Monetization Algorithm
‘Big YouTube news is breaking today and so many creators are going to be fucking pissed’
YouTube’s creator base has been in informal conflict with the platform that hosts their videos for years. And now, they think they have evidence of just how the platform is screwing them over — by prioritizing family-friendly videos from mainstream outlets over the work platform creators when it comes to monetization.
For the last month, a handful of creators have been probing a vulnerability they believe they’ve uncovered that has inadvertently leaked an internal score YouTube uses to rate their channel and videos: the P-Score.
“Our proprietary algorithm, the P-Score, looks at the popularity and viewer passion of specific content — things like the amount of repeat views and how often videos are shared,” said Kate Stanford, YouTube’s head of global advertiser marketing, at the company’s Brandcast event this May.
A video’s — and a channel’s — P-Score (short for preference score) dictates whether or not content is surfaced in YouTube’s Google Preferred lineup, the advertiser-friendly, vetted, high-quality content that YouTube introduced back in 2013 and tightened the criteria for in 2018, post-Adpocalypse — the incident in spring 2017 where huge advertisers pulled their cash from YouTube after their products were being advertised against terrorist propaganda. Those in the Google Preferred lineup are just that — preferred, and able to access more advertising opportunities, and a higher return on any adverts sold against their videos.
A channel’s P-Score is graded using five key signals, according to YouTube: popularity, driven by watch time and engagement; passion, which measures channel engagement; protection, which ensures that content is appropriate; platform, which highlights content watched often on TV screens; and production, which measures the production values of a video.
Videos in the Preferred silo are first checked using machine learning, then are manually vetted — providing brands looking to advertise on YouTube with the assurance that they’re not going to appear against inappropriate content in the same way they did in the Adpocalypse.
The goal of P-Score, which ranks the top 5% of content popular with the key 18–to-34 demographic most important to advertisers, is to provide “engaging, brand appropriate content on YouTube”.
The P-Score was discovered in the source code of each video hosted on YouTube, and is believed to have surfaced following a change to YouTube on or around October 8th. Many high ranking YouTube channels that frequently feature in YouTube’s Trending section are given scores in the high hundreds.
Various groups of creators have been working through millions of channels’ P-Scores over the last few weeks, trying to reverse engineer how it works, and what it means for them as creators trying to make a living on the platform.
Teams have been secretly sifting through huge corpuses of videos in order to try and gather as much data as possible about how YouTube’s P-Score algorithm works without tipping off YouTube that the data was publicly accessible. Some began tweeting about it this afternoon.
Amongst their purported findings are that a channel’s P-Score can differ depending on where the viewer is watching. A German YouTube viewer accessing WatchMojo, a major YouTube channel which has 37 million subscribers who spend close to two billion minutes of watch time on YouTube each month, will see the channel given a P-Score of around 755, while an American viewer accessing the same channel will see it given 923.
“As one of YouTube’s longest standing partners and most successful native digital brands on the platform, we recognized early on that serving audiences’ interests by informing and entertaining them while respecting their time was key to building a successful community — which then leads to likes, comments, and shares,” says Ashkan Karbasfrooshan, founder of WatchMojo.
“As an engineering-based culture, it’s not surprising to see YouTube’s team always come up with metrics and systems to quantify, rank, measure and make sense of different channel’s quality and engagement.”
Those who have gone through the data believe that a well-established, average YouTube channel with a high profile will rate between 600 and 700 or so, and estimate that those in Google Preferred — the top 5% of channels — score higher than around 800.
Several groups of YouTubers have been combing through the data to try and discern patterns. One group, including YouTubers Bowblax, Nicholars DeOrio, Optimus and Josh Pescatore, have analyzed more than 200 randomly selected large YouTube channels to ascertain their P-Score.
The highest-ranked brand-friendly YouTubers are who you would expect — the late night TV hosts beloved by the platform, plus some news outlets and Will Smith, the face of YouTube’s much-maligned 2018 Rewind video.
PewDiePie is the 13th-highest ranked YouTuber based on his P-Score — perhaps a surprise given his chequered past and worries about being shunned as the face of the platform due to controversial videos he has posted.
“I think this confirms our long time suspicions that ‘homegrown talent’ is being pushed aside in favor of ‘advertiser friendly’ late night TV hosts,” says YouTuber Nicholas DeOrio. “It’s a concept that a lot of people had assumed was true already however, actually seeing the numbers being this skewed is still pretty disappointing.”
Those looking through the data have also discovered references to “throttling” in the source code of video pages, which they believe could indicate YouTube limiting traffic to their videos.
YouTube has been approached for comment for this story, which will be updated with their response.
But for the creators who have pored over the data, their findings are vindication of something they’ve believed and feared for a while.
“Creators on the platform have long worried that we’ve had to deal with demonetization not only affecting our income, but video performance,” says YouTuber Optimus. “When we look at our analytics, you can literally see the decline in viewership the minute that videos become demonetized. It’s been sort of a conspiracy, and it’s something that YouTube has denied publicly.
“It doesn’t seem like they lied in literal terms, but when you look at the code, there appears to be two types of throttling: ad throttling and feature video throttling. Ad throttling constantly refers to certain types of ad spaces being rejected for sale or the video not being brand safe or even the video not being allowed certain types of experimental ads,” says Optimus.
“Video throttling seems to be more algorithmic because not only does it explain why we notice these decreases so often, but also it’ll constantly include reasons like the video not being family friendly, the video being throttled for ‘overlay’, etc.”
The creators’ work rifling through P-Scores is vindication of their long-held belief, says Optimus: “If anything with all of this, we’ve just shown that YouTube creators have been right this whole time.”
But Karbasfrooshan cautions against reading too much into a channel’s P-Score: what matters first and foremost is producing high-quality content. “While we naturally welcome having a higher P-score than a lower one, it’s critical to remember that successful editorial isn’t limited to science and metrics alone, but resonating with the emotions and feelings that make a viewer watch one channel over another,” he says.