On 24 November 2024 four Chinese authorities launched the “Qinglang · Algorithmic Governance” special campaign, which ran until 14 February 2025. Qinglang campaigns are a familiar fixture, but this one is unusual: it targets the algorithm itself rather than the content the algorithm surfaces.
What the campaign targets
The campaign identifies three broad harms. The first is the filter bubble — the tendency of recommendation systems to narrow what users see and to homogenise public discussion. The second is manipulated rankings: hot-search lists, trending topics and recommendation slots that can be bought, gamed or otherwise distorted. The third concerns gig workers, where algorithmic scheduling and rating systems are said to push riders and drivers toward unsafe workloads.
Together, the three describe a single concern: algorithms as a mechanism of influence over people, whether users or workers. That is a step beyond the older focus on individual pieces of illegal content.
How it connects to existing rules
The campaign is enforcement, not new law. It operationalises obligations that already exist in the measures on internet information services with intelligent services, which require transparency about algorithmic recommendation and give users a way to opt out of personalised push. The campaign also sits within a well-established Qinglang line of work: the AI cleanup phase-one guide and the AI misuse phase-one progress report show how these campaigns are sequenced, while the later 2025 AI abuse campaign and the self-media AI labelling effort extend the same machinery to synthetic content.
Read as a set, they show that Qinglang is a recurring instrument with recurring targets, and that each campaign tends to leave behind permanent expectations even after the campaign window closes.
Practical actions for AI companies
First, audit your recommendation logic. If the system can be shown to narrow content systematically, that is now an identified enforcement target, and a documented review is better than a surprised one.
Second, keep the user-facing opt-out working. The right to disable personalised recommendation is an existing requirement, and it is one of the easiest things for a regulator to test.
Third, review ranking and promotion products. Any feature that lets a paying party move content up a list should have a clear label and an auditable rule.
Fourth, examine worker-facing scheduling tools. If your product allocates tasks or scores performance, the human consequences of those choices are now explicitly in scope.
None of this requires rebuilding a product. It requires knowing which of these four areas your system touches, and being able to explain the answer.
This article is based on the announcement of the ‘Qinglang · Algorithmic Governance’ special campaign and public information and is for reference only; it does not constitute legal advice.