Creator Platforms Don't Govern Creators. Their Networks Do.

Based on the research ofQing Xiao et al., "Constructing Algorithmic Authority: How Multi-Channel Networks (MCNs) Govern Live-Streaming Labor in China," arXiv, 2026

A nine-month ethnographic study of 44 live-streamers and multi-channel network (MCN) staff in China, by Qing Xiao and eight coauthors, asks who actually governs creator labor. The answer is not the platform. On Douyin, Taobao Live, and their peers, streamers rarely touch the recommendation algorithm directly. They deal with an MCN that signs them, takes a cut of their revenue, and tells them how to perform. Xiao's fieldwork shows these intermediaries manufacture what the authors call algorithmic authority: they hold a probabilistic view of the algorithm for themselves while selling talent a deterministic one. We call the space between those two accounts interpretive arbitrage, and it is where the intermediary's margin lives.

Consider what that layer is worth. On March 24, 2014, Disney announced it would pay $500 million, and as much as $950 million with performance earnouts, for Maker Studios, an MCN that had signed more than 55,000 YouTube channels reaching a combined 380 million subscribers. Disney was not buying content. It was buying an interpretation layer: a company that claimed to know how to read YouTube's algorithm on creators' behalf and translate that reading into scale. Twelve years later, Xiao's study documents the same business model at industrial scale, and shows exactly how it makes its money.

Platforms like Douyin, Taobao Live, and YouTube publish rules, but rules are not the same as governance. In practice, complementors, the streamers and sellers who generate the content, rarely deal with the algorithm directly. They deal with the MCN, and that intermediary layer is where the actual governing happens. Xiao's research shows it runs on a deliberate asymmetry: a probabilistic account of the algorithm kept for internal risk management, and a deterministic one sold to the people whose income depends on it. The gap between those two accounts shows up in three places: the take, the story, and the unwind.

The Take

MCNs are not paid a flat management fee. They are paid a share of a commission stream that only makes sense once you assume the algorithm is legible enough to optimize. On Taobao Live, gross sales commission typically splits roughly 1:2:7 among Alibaba, the Taobao Live platform, and the MCN itself, meaning the intermediary that never appears in the marketplace's terms of service collects seven times what the marketplace does (see the exhibit). On Douyin, brands pay around 20% of gross merchandise value in commission; after the platform's own cut, the remainder is split roughly 50/50 between the MCN and the streamer. Douyin's e-commerce business generated close to 3.5 trillion yuan (about $487 billion) in 2024, and by 36Kr's count, top-tier influencers accounted for only about 9% of that volume. Most of the value, and most of the commission, runs through mid-tier accounts that depend entirely on their MCN's reading of the algorithm to know where to invest their next hour of airtime.

That dependence scaled into an industry. The number of Chinese MCNs exploded from fewer than 2,000 in 2017 to nearly 5,000 in 2018, and then to roughly 28,000 by 2020, according to Qianzhan and iiMedia Research (see the exhibit). Each of those agencies exists to sit between the platform and the talent, and each takes its cut for reading an algorithm none of them controls.

The Story

Value capture at that scale needs a justification, and Xiao's fieldwork found the justification is manufactured differently for two audiences. Internally, MCN managers treat the algorithm the way a hedge fund treats a volatile market: uncertain, probabilistic, to be hedged against with diversified rosters and constant experimentation. Externally, in conversations with streamers, the same managers circulate a simplified narrative in which the algorithm is transparent, fair, and responsive to individual effort.

Underperform, and the story says to stream longer, smile more, or push a sharper script, never that the model recommending your content changed weights last Tuesday for reasons the MCN itself cannot fully explain. We read that double narrative not as a communication failure but as the mechanism that makes a 70% commission tolerable: a streamer who believes the algorithm rewards effort will accept a smaller share of the outcome, because the outcome still feels like theirs to improve.

A streamer who believes the algorithm rewards effort will accept a smaller share of the outcome.

The Unwind

The arrangement holds only as long as the external story keeps roughly matching lived outcomes, and Maker Studios shows what happens when it stops. By 2017, Disney had shrunk the network from tens of thousands of channels to about 300, after growth failed to justify the earnout Disney had priced in; it cut roughly 80 staff and folded what remained into its Disney Digital Network unit. The certainty Maker had sold Disney about algorithmic reach turned out to be no more reliable than the certainty MCNs sell individual streamers today.

In China, the same unwind shows up at the top of the distribution first. Streamers with enough leverage break from their MCNs to build in-house studios once they conclude the platform relationship itself, not the MCN's reading of it, is the asset worth owning. Austin Li Jiaqi, the "Lipstick King" with more than 60 million Taobao followers, operates through his own company, Meione, rather than renting another agency's interpretation of the feed. What keeps the mid-tier majority in place is the same interpretive gap that let Maker scale in the first place, now running in reverse: the less legible the algorithm actually is, the more a confident-sounding intermediary is worth, right up until its confidence is tested against results.

The Rule for Platform Owners

If you run a creator platform, the lesson is not that MCNs should be regulated out of existence; intermediaries that translate opaque systems for the people who depend on them are a real service. The lesson is that governance built through an interpretive layer is governance you do not fully control and often cannot audit, because you never see which of the two stories a given MCN is telling. Douyin and Taobao Live have both experimented with transparency dashboards and direct-to-streamer performance data for exactly this reason, not to eliminate MCNs but to shrink the gap that interpretive arbitrage depends on. Narrow that gap without dismantling the intermediaries your complementors still need, and you start governing your ecosystem directly instead of governing it by proxy.