Chinese regulators have not banned GEO. Since 2026 they have treated one form of it as AI data poisoning , an enforcement target. That form is feeding models fabricated content so that they recommend a product. This entry lists the documents that matter and when each one appeared. The Chinese sources are summarised here; the Chinese version quotes them.
Date What Type 2023-08-15 Generative AI interim measures take effect Regulation (CAC and six ministries) 2025-09-01 AI-generated content labeling measures take effect Regulation (CAC and others) 2026-02 AIIA’s “AI safety commitment: GEO” pledge signed at the alliance’s 16th plenary by 10 companies Industry self-regulation 2026-03-15 CCTV’s 3·15 consumer-rights show exposes GEO poisoning Media investigation 2026-04-30 CAC launches the four-month “Qinglang: rectifying AI application disorder” campaign Enforcement campaign 2026-05-29 CAC and four other departments publish the provisions on multi-channel distribution services for internet information content, in force 2026-09-01 Regulation 2026-09-14 CAAC publishes five GEO group standards, T/CAACCHINA 001–005-2026 Voluntary group standard
The CAC’s campaign runs for four months in two stages. Stage one targets AI services. Its fourth of seven problems is AI data poisoning: tampering with training corpora, forging authoritative data, or using GEO for malicious marketing. Selling or teaching poisoning methods and tools is also covered. The same item faults models that don’t cross-check cited sources, give no risk warnings, or don’t link the information they cite.
Officially documented · 新华网（中央网信办通知）
Stage two targets content: AI-generated low-quality “digital slop”, fabricated information and impersonation. It also targets AI-run fake-account networks that mass-produce near-identical posts and fake engagement. Platforms, accounts and MCN agencies that break the rules are to be dealt with and penalised.
Officially documented · 新华网（中央网信办通知）
Mass-producing near-identical “recommendation” articles across accounts touches both stages. The content itself is a target, and so is the poisoning of what models cite.
On 29 May 2026 the CAC, the Ministry of Public Security, the Ministry of Culture and Tourism, the State Administration for Market Regulation and the National Radio and Television Administration announced the Provisions on Multi-Channel Distribution Services for Internet Information Content, in force from 1 September 2026. The CAC’s Q&A defines the service as planning, production, distribution, marketing, promotion and brokerage services for the content that users publish through public accounts. Its list of prohibited conduct includes fabricating topics or forging and splicing content to mislead the public, inventing backgrounds, plots or personas for false or misleading marketing, and faking attention, views, clicks or ratings or publishing homogeneous content in bulk.
Officially documented · 国家互联网信息办公室
The Q&A does not mention AI or GEO. The order number and the penalty clauses are in the full text, which this entry has not checked. Whether and how the provisions apply to a GEO provider is not stated there. The definition covers services that plan, distribute and promote content published through public accounts, so ask a provider whether it falls within scope.
CCTV’s 3·15 show reported that some providers mass-produce false information to steer AI answers. In the test it described, a GEO tool generated and published more than ten promotional articles about a fitness band that did not exist. Two major AI models then recommended it near the top.
Observed in practice · observed 2026-03 · 第一财经
In force since 1 September 2025. AI-generated content carries explicit labels people can see, and implicit labels in file metadata. Users who publish AI-generated content must declare it. Nobody may maliciously remove, alter, forge or hide labels, or provide tools for doing so.
Officially documented · 国家互联网信息办公室
The generative AI interim measures have been in force since 15 August 2023. They forbid generating false and harmful information that the law prohibits. Providers must make services more transparent and generated content more accurate and reliable.
Officially documented · 国家互联网信息办公室
These are voluntary, not law. They matter because buyers use them to judge providers.
The China Advertising Association of Commerce (CAAC) published a five-part GEO group standard on 14 September 2026. It had passed review on the national group standards platform. The five parts are: general principles and terms, provider evaluation, pricing and measurement, trusted corpora, and process compliance and safety.
Observed in practice · observed 2026-09 · 界面新闻（东方财富转载）
The China AI Industry Alliance (AIIA) initiated a GEO pledge (titled, in translation, AI Safety Commitment: GEO), signed at an alliance plenary. Its safety governance committee was also drafting a technical specification, Trusted Requirements for GEO Services (translated title).
Observed in practice · observed 2026-06 · 中华网
AIIA’s own report dates the signing: the pledge was “formally signed in February 2026 at the 16th plenary of the China AI Industry Alliance” (quote translated from Chinese). The first signers were 10 GEO-related companies, including Qinglan Zhihui, BlueFocus and Zhihu, and nearly 20 companies had joined since. The report says more than 20 organizations contributed to the Trusted Requirements for GEO Services specification. The China Academy of Information and Communications Technology (CAICT) has launched the first round of assessments against it. The China.com report from June came after the signing. This entry previously dated the pledge “by June 2026” on that basis.
Observed in practice · observed 2026-05 · 中国人工智能产业发展联盟（AIIA） (unpublished report, not publicly available; readers can't check it)
Reports often cite draft versions of the CAAC standard, for example minimum sample sizes for measuring visibility. Check figures against the published text; drafts change.
Seeding fake reviews to influence AI answers falls under existing consumer protection law in the US.
It is a violation “for a business to write, create, or sell a consumer review, consumer testimonial, or celebrity testimonial that materially misrepresents, expressly or by implication” that the reviewer exists or used the product, or their experience with it. The rule also covers buying reviews, insider reviews, company-controlled review sites, review suppression and fake social media indicators.
Officially documented · US Federal Trade Commission (eCFR)
Publishing accurate information about your own products, in places models read, is not the target. The target is fabricated products, reviews, rankings or “authoritative” data.
A provider that sells guaranteed AI rankings through mass-produced articles is offering what the campaign calls poisoning. See Claim: seeding brand mentions across the web boosts AI visibility .
Label AI-generated content you publish in China.
Google’s spam policies cover attempts to manipulate generative AI responses. See Google spam policies .