AI Creating for AI? Inside the Strange Cycle Emerging in Short Videos

Racking up views through AI-Powered Watching

By now, most people have probably watched YouTube Shorts. They provide instant gratification in a very short amount of time, and because the next video starts playing immediately, you can end up spending a surprising amount of time watching them without realizing it. The same goes for Instagram Reels and TikTok, but short-form videos have become one of the things that consume huge amounts of modern people’s time.

What we call short videos are often created by taking a longer, original video and cutting out only the parts that are likely to attract the most viewers.

These clipped short videos are often made to direct viewers to the original YouTube video or other source, but sometimes the sole purpose is to rack up numbers such as views and subscribers. Recently, it has also become easy to create videos using generative AI and add AI-generated voiceovers. And now, there are even cases where AI is used to watch the videos themselves in order to inflate view counts.

In 2024, a man was arrested on charges including wire fraud after allegedly having AI bot accounts play hundreds of thousands of AI-generated songs he had uploaded billions of times, earning approximately ¥1.4 billion in royalties. Similar methods are reportedly being used to inflate video view counts as well.

“AI-based view inflation works by instructing systems such as ‘View Bots,’ which automate browsing, to increase the number of times a video is played. Bots model human viewing behavior, and in recent years they have become capable of mimicking behaviors such as watch time, where the video is stopped, seeking through the video, scrolling speed, and the intervals between clicks.

“Platforms such as YouTube are also taking measures against fraudulent manipulation, so operators are trying to get around those safeguards by making the bots behave more like humans in order to rack up views.

“There are also systems that combine this with browser automation. This is technology that reproduces and automatically executes actions that humans would normally perform in a browser. It has evolved to the point where it can carry out an entire sequence of actions, such as opening a browser and searching for a video and watching related videos.” — IT writer

It seems that platforms are also continually evolving their countermeasures against such fraudulent attempts to generate views.

“Platforms analyze user sessions and determine whether activity is legitimate based on a wide range of factors, including the device, network, and user behavior. Simply making the activity look human isn’t enough to avoid detection, so the operators develop increasingly sophisticated technologies to evade monitoring. It has become an ongoing cat-and-mouse game.” — Same source

The method is spreading through production manuals

Short videos produced using AI generally make use of videos that have gone viral on social media, with overseas videos being particularly common. Recently, around the end of December last year, a video titled something along the lines of “What a former child actor from a popular overseas drama looks like now as a homeless person” spread across social media and attracted considerable attention.

From there, explanatory videos were created on YouTube, Instagram, and other platforms, resulting in viral content being recycled across the internet.

The topics commonly used for these short videos are things that grab people’s attention, such as everyday questions or strange and mysterious phenomena. AI is then used to create explanatory videos based on those topics, but the explanations are often incorrect as well. Content that can hardly be considered legitimate material is simply being churned out and distributed on YouTube.

Why have these types of short videos spread so rapidly? I spoke with writer Yusuke Yamano, who is knowledgeable about AI-based image and video editing, and he pointed out that one factor is that the methods for producing them have rapidly spread through paid information products on platforms such as note.

“For example, if you search for ‘AI short videos’ on note, you’ll find countless guides explaining how to make them, ranging from pages that can be viewed for free to paid content. They include instructions such as using ChatGPT to create the script and AI-generated voices for the narration. The goals of the creators vary—from attracting an audience through short videos to directing viewers to LINE groups or their own paid note content.”— Yamano

Why, then, have so many overseas videos come to be circulated as short-form content? One reason, Yamano says, is that it’s harder for the people who own the rights, such as the person who filmed or originally posted the video, to discover them.

“Short videos made from clips are generally created without permission, so if the rights holder discovers them, there is a possibility that they will request their removal. If the clips are edited for a Japanese audience, YouTube’s algorithm makes them less likely to reach people overseas. The idea that this was a relatively safe way to create pirated content spread, and it probably became the standard. There are huge numbers of clips from Japanese television programs circulating overseas.” — Yamano, and hereafter

A tactic for making money by selling accounts

Apparently, the most reliable way to make money from mass-produced short videos is to sell the YouTube account itself through websites dedicated to buying and selling YouTube accounts.

“There are many sites where YouTube accounts can be bought and sold, such as M&A Club and Rakkō M&A. The price is determined by factors such as how many subscribers the account has and what kind of videos it contains. Even accounts with fewer than 10,000 subscribers can sometimes be sold for ¥100,000 to ¥200,000.

“What matters is how much revenue can be expected after purchasing the account. With videos featuring a human presenter, changing the person appearing in the videos as a result of the sale could cause views and revenue to decline. That’s why accounts that use AI-generated voices and minimize dependence on any particular individual are preferred.”

Everything from writing the video scripts to editing can be done using AI. There are even imitator creators who make videos that are virtually identical to viral short videos. Yamano also says, “There are people selling scripts designed for AI creators, but those scripts are AI-generated too. AI-generated material is fed into AI to create videos. It’s a vicious cycle, which is why so much of the content is low quality.”

Every day, videos created solely to rack up views are being uploaded online. Videos are made with AI to generate numbers, then those videos are further edited into short-form content by AI, and in some cases, even the videos themselves are watched by bots to inflate their view counts. From AI-driven production and editing to automated viewing, a bizarre cycle is beginning to emerge in which “AI consumes content mass-produced by AI” without any human involvement.

At the 13th Nikkei “Hoshi Shinichi Award,” announced in February, three of the four winning works in the general category—including the Grand Prize winner—were created using generative AI. Even the judges commented that they were completely indistinguishable from works created by humans.

The distinction between what is created by AI and what is created by humans is becoming increasingly blurred. Finding the right way for humans to coexist with AI will be a major challenge going forward, but online, the pollution caused by AI-generated content is already spreading on a massive scale.