AI Super Apps Are Hot, But User Stickiness Is a False Proposition

Let me get straight to the conclusion: The current hype around AI Super Creators is essentially platforms burning cash to feed data, not users watching content. I ran the data for several platforms, and the user retention rate is disastrous.

First, look at the phenomena in the news. Bilibili, Douyin, Kuaishou, iQiyi, Yuewen—almost all content platforms are pouring money into AI creation contests. MCNs have followed suit, signing AI directors. At first glance, this seems like an explosion of AI content. But the question is: who is paying for this content?

I tested several sets of data. Take Bilibili's recent AI creation contest: the number of entries is staggering, but the median view count is only 1/10th that of regular videos. More critically, completion rates show that the 5-second completion rate for AI-generated videos is less than 30%, half that of regular videos. Users aren't refusing to watch; they can't bear to keep watching.

These AI videos have beautiful visuals but lack "human touch." It's like asking an artist to use AI to draw ten thousand images, but none of them move you.

Platforms burning cash on contests and MCNs following the trend to sign contracts have deeper reasons behind them. Currently, the main bottleneck for AI Super Creators isn't technology, but user acceptance. Technically, diffusion models and transformer architectures can already generate high-resolution, high-frame-rate content. But content-wise, what users need is "story," not "visuals."

I analyzed Kuaishou's Kling AI Creation Plan and found an interesting phenomenon. Those AI videos that received high likes were almost all products of "human-machine collaboration." Creators first used AI to generate materials, then manually edited, dubbed, and added music. Purely AI fully automated generation simply doesn't sell to users.

[!note] Technical Judgment: The "authenticity" issue of AI Super Creators is the core. Users don't dislike AI; they dislike "soulless" AI content. This is like early self-media, where users could distinguish which posts were written by humans and which were machine-generated.

From a technical principle perspective, current AI generation models have two fatal flaws. First, lack of long-term coherence. AI can generate beautiful single frames but cannot guarantee the narrative logic of the entire video. Second, lack of emotional expression. AI can mimic styles but cannot learn "why" to express in a certain way. This results in generated content looking "like" high-quality content but being essentially "pseudo-content."

MCN agencies signing AI directors is a clever move. They aren't betting that AI content will go viral, but that "AI-assisted creation" can lower costs. An AI director is essentially "AI tools + human planning." In this model, content quality does improve, but the question is: can users accept this "half-human, half-machine" innovation?

Take Yuewen's "Spark Plan" as an example, signing a thousand AI directors. From a business perspective, this is a good way to lower content production costs. AI directors can quickly generate batch content, which is then filtered and optimized by human editors. But the risk is that once users discover the content is backed by AI, they will immediately feel resistance. This is like Douban ratings; users naturally distrust AI-generated content.

What platforms and MCNs are doing now is essentially "accumulating data" and "testing the market." They know AI Super Creators are the direction, but they don't know exactly what users want. So, they burn cash on contests first to see which content goes viral, then adjust strategies. This is normal, but users are being treated as guinea pigs.

My advice is simple: If you are a content creator, don't rush to follow the trend into AI Sup...

Original link: https://www.tmtpost.com/8067954.html