More than 20% of the videos that YouTube's algorithm shows to new users are “AI junk” – low-quality AI-generated content designed to drive views. the study showed.
Video editing company Kapwing examined 15,000 of the world's most popular YouTube channels (ranked in the top 100 in every country) and found that 278 of them only contained Backup slop.
These AI-powered channels are estimated to have amassed more than 63 billion views and 221 million subscribers, generating around $117 million (£90 million) in revenue each year.
The researchers also made a new YouTube account and discovered that 104 of the first 500 videos recommended on his feed were AI garbage. A third of the 500 videos were “brain” videos, a category that includes AI crap and other low-quality content created to monetize attention.
The results provide a snapshot of a rapidly expanding industry that is saturating major social media platforms – from X to Meta to YouTube – and defining a new era of content: decontextualized, immersive and international.
Guardian analysis this year, it was revealed that nearly 10% of YouTube's fastest-growing channels were AI duds that racked up millions of views. despite the platform's efforts to limit “inaccurate content.”
Channels discovered by Kapwing are distributed and viewed worldwide. They have millions of subscribers: in Spain, 20 million people, or almost half the country's population, follow trending AI channels. AI Channels has 18 million subscribers in Egypt, 14.5 million in the US and 13.5 million in Brazil.
Bandar Apna Dostthe most-watched channel in the study is based in India and currently has 2.4 billion views. It follows the adventures of an anthropomorphic rhesus monkey and a muscular character modeled after the Incredible Hulk, who fights demons And travels in a tomato helicopter. Capwing estimates the channel could earn up to $4.25 million. Its owner did not respond to a Guardian request.
Rohini Lakshane, a technology and digital rights researcher, said Bandar Apna Dost's popularity likely stems from its absurdity, hyper-masculine imagery and the fact that it lacks a plot, making it accessible to new viewers.
Pouty Frenchmanbased in Singapore, has 2 billion views and appears to be aimed at children. It follows the adventures of a French bulldog – a trip to a candy forest, eating crystal sushi – many of them accompanied by children's laughter. Capwing estimates she earns about $4 million a year. Fascinating storiesbased in the US, it also targets children with cartoon themes and has 6.65 million subscribers, making it the channel with the most subscribers in the study.
Meanwhile, World of AIbased in Pakistan, features AI-generated short films about the disastrous floods in Pakistan with titles such as “Poor People”, “Poor Family” and “Flooded Kitchen”. Many of these videos are accompanied by a soundtrack called Relaxing Rain, Thunder & Lightning Ambience for Sleep. The channel itself has 1.3 billion views.
It's hard to say exactly how significant these channels are compared to the vast sea of content already posted on YouTube. The platform does not publish information on how many views it has annually or how many of them are due to AI content.
But behind these eerie scenes of candy forests and disasters lies a semi-structured, growing industry of people trying to find new ways to monetize the world's most powerful platforms using artificial intelligence tools.
“There are a lot of people on Telegram, WhatsApp, Discord and message boards exchanging tips and ideas. [and] sell courses on how to make slop that will be interesting enough to make money,” said Max Reedjournalist who has written extensively on the topic of artificial intelligence.
“They have what they call niches. I recently noticed AI videos of people blowing up pressure cookers on the stove.”
While AI misfit creators are everywhere, Reed said many of them come from English-speaking countries with relatively strong internet connections, where the average wage is less than what they can earn on YouTube.
“Mostly middle-income countries like Ukraine, lots and lots of people in India, Kenya, Nigeria, quite a few in Brazil. You also see Vietnam. Places with relative freedom of online access to social networking sites,” he said.
Being a creator of AI trash isn't always easy. First, YouTube and Meta's creator programs aren't always transparent about who they pay for content and how much, Reed said. On the other hand, the AI garbage ecosystem is full of scammers: people selling tips and courses on how to create viral content, who often make more money than the AI garbage producers themselves.
But at least for some it's a livelihood. And while new, attention-grabbing ideas like exploding pressure cookers emerge all the time, when it comes to AI dregs, human creativity matters far less than the algorithms that distribute content to Meta and YouTube.
“These websites, by their nature, are huge A/B testing machines,” Reed said. “Almost everything you can think of, you can already find on Facebook. So the question is, how do you find the things that are doing well and then how do you scale them up? How do you make 10 of them?”
A YouTube spokesperson said: “Generative AI is a tool, and like any tool, it can be used to create both high-quality and low-quality content. We remain focused on delivering high-quality content to our users, no matter how it was created. All content uploaded to YouTube must adhere to our community guidelines, and if we find that content violates the policy, we will remove it.”
Quick Guide
Contact us about this story
Show
The best public interest journalism relies on first-hand accounts from people who know.
If you have something to share about this topic, you can contact us confidentially using the following methods.
Secure messaging in the Guardian app
The Guardian app has a tool for submitting story tips. Messages are fully encrypted and hidden as part of the routine activity that every Guardian mobile app performs. This prevents the observer from knowing that you are communicating with us at all, let alone what is being said.
If you don't already have the Guardian app, download it (iOS/Android) and go to the menu. Select Secure Messaging.
SecureDrop, instant messengers, email, phone and mail
If you can safely use the Tor network without being tracked or monitored, you can send messages and documents to the Guardian through our SecureDrop Platform.
Finally, our guide to theguardian.com/tips lists several ways to contact us securely and discusses the pros and cons of each.
