Why China's internet evolved separately

Platforms everywhere operate inside a legal environment. The EU has the GDPR and the Digital Services Act. The US has Section 230 and a patchwork of state laws. China's differs in content and emphasis, not in the fact of existing, and the practical question for anyone studying the market is simply: what do the rules require, and what shape does that produce?

This lesson answers that descriptively. The aim is to explain the mechanisms and their structural consequences, the same way the earlier lessons explained retrieval and ranking, without arguing about whether any of it is desirable.

The short version is that operating a consumer platform in the mainland involves obligations that are, in aggregate, more prescriptive and more operational than in most Western markets: you need permission to operate, accounts are tied to verified identities, platforms carry responsibility for what is published, and some algorithmic systems must be registered with a regulator.

Each requirement has a knock-on effect on product design. Those effects are the actually interesting part, and they explain much of why the two ecosystems look so different.

Two foundational requirements shape everything above them.

Licensing. Running a website or online service for users in the mainland requires an ICP licence or filing, issued by the telecoms regulator and tied to a registered local entity. Hosting providers check it. This makes operating a platform an activity you are permitted to do rather than one you simply start, which is a meaningful difference from a market where anyone can register a domain and launch.

Real-name registration. Accounts on major services must be linked to a verified identity, in practice a mobile number, itself registered to an ID. Users still post under nicknames, so it is not public identity; it is accountability behind the screen name.

The product consequences follow mechanically. There is effectively no throwaway-account layer, which changes the economics of spam and bot networks: fake accounts require real phone numbers, so they cost money rather than nothing. Age is knowable, which makes minors' rules enforceable. And identity is available for payments, which is one reason commerce integrated so smoothly into social apps.

The single most consequential difference for how these products are built is where responsibility sits.

In the US, Section 230 broadly shields a platform from liability for user posts. That immunity is why American platforms could scale to billions of users with comparatively small moderation staffs: the law did not require them to inspect everything, so the economics permitted them not to.

In China the expectation runs the other way. Platforms are responsible for content they distribute, which makes moderation a licence-preserving function, not a cost centre to be minimised.

The structural results are visible in the products:

• Moderation is in-house and large, combining automated classifiers with substantial human review, rather than being outsourced as thinly as possible.
• Review often happens before or at publication for some content types, not only after a complaint.
• Compliance sits inside product and engineering, shaping what ships, rather than reviewing it afterwards.

It is the same causal chain the moderation lesson describes: where liability sits determines how much reviewing gets done. Two legal regimes, two staffing models, two products.

Here is a requirement with no close Western parallel, and it is genuinely interesting as engineering policy.

Since 2022, providers using algorithmic recommendation for users in the mainland must file details of those algorithms with the Cyberspace Administration of China, which maintains a public registry of filed systems. The rules also require, among other things, that users can turn personalised recommendation off, and that providers do not use algorithms to unreasonably differentiate pricing between users.

Set aside the politics and notice what this does technically. It treats a ranker as a regulated system with a filing obligation, similar to how a financial model or a medical device is treated: you must be able to describe it, and an authority holds a record of it.

Compare with the EU's Digital Services Act, which pushes at the same object from a different angle by requiring very large platforms to explain recommender parameters, offer a non-profiling option, and submit to risk audits.

The convergence is worth noticing: two very different jurisdictions independently concluded that recommendation is infrastructure worth regulating. The mechanisms differ; the instinct is the same.

Because age is knowable, rules for minors are enforceable rather than aspirational, and that produces something the West largely does not have: a separate product for under-18s.

Major platforms operate a youth mode or minors mode, which is not simply a content filter. It typically ships a different recommendation feed built around educational and age-appropriate material, plus time limits on daily use and curfews overnight. Gaming carries its own well-known restrictions on play time for minors.

The mechanism worth extracting is the dependency chain. Verified identity makes age reliable. Reliable age makes an under-18 experience enforceable. Enforceability makes a genuinely separate product worth building.

Western platforms have moved in a similar direction with teen accounts and default protections, but they run into the same wall each time: without reliable age, every control is a self-declared checkbox a determined teenager clicks through. Age assurance is the live debate in the EU and UK precisely because it is the missing link.

Same goal, different starting point, different achievable product.

The clearest evidence that a separate ecosystem grows separate things is the micro-drama, or duanju.

The form: a scripted series told in vertical episodes of roughly one to two minutes, running to dozens or hundreds of episodes, each ending on a cliffhanger. Plots are unapologetically broad, revenge, sudden fortune, secret identities. Production is fast and cheap.

The scale is the surprise. CNNIC reported the micro-drama audience at 662 million as of December 2024, close to 60 percent of China's internet users. That is not a niche; it is a mass medium that assembled itself in a few years.

The economics are what make it structurally novel: the first episodes are free, then viewers pay per episode or unlock a series, often through a mini-program rather than a media app. So it is closer to mobile-game monetisation applied to television, and it inverts the ad-supported model that funds Western short video.

Why no Western twin? The same reason as livestream commerce: it needs frictionless micro-payments already sitting in the app. Ask a viewer to enter a card at the cliffhanger and the format dies.

Pull the four lessons together and the divergence is not a list of quirks. It is one causal chain.

Read top to bottom: payment inside the app makes commerce native, which makes creators sellers, which makes livestreaming and micro-dramas viable, which makes private traffic worth owning. Verified identity makes minors' products enforceable. Platform liability makes moderation a core function.

Several widely used foreign platforms are not available in the mainland, so domestic services filled each role and then evolved under these conditions for a decade. The result is not the Western internet with different logos. It is a different set of answers to the same questions, and reading it as behind or ahead misses the point entirely: it went sideways, not forward or back.

The differences are downstream of a few starting conditions. Payment inside the app drives the commerce shape; verified identity drives the minors and spam picture; platform liability drives the moderation build.