China's GEO Standard Graded Every Source A to D — Overseas Brands Start at C

For an overseas brand, the Chinese AI-search game has always felt like it was played by someone else's rules. Now it literally is. On August 11, China published its first credible-communication standard for generative engine optimization (GEO), and it took effect the next day.

The standard — T/CAPT 026-2026, the Generative Engine Optimization (GEO) Credible Information Communication and Information Ecosystem Governance Specification — does not tell brands how to rank. It does something sharper: it grades every source an AI model might cite into four tiers, A through D. Government records and state-media originals sit in tier A. Academic papers and association white papers sit in tier B. Corporate websites — the one asset nearly every overseas brand starts with — sit in tier C, flagged as "auxiliary" material that needs extra verification for any core factual claim.

That one sentence changes the economics of a China GEO program. If your brand only exists at tier C, no amount of content volume will carry you into an AI answer with the same authority as a tier-A or tier-B source. The standard was drafted by Xinhua-affiliated laboratories, not by marketing vendors, and it is already being read as a reference for enforcement. Here is what an overseas operator needs to understand, and what to do about it.

✅ The Standard Is Already in Force

Three dates matter. The draft passed its expert review on August 9, was published on August 11, and took effect on August 12. On August 27 it was re-released as a forum outcome at HICOOL 2026, Beijing's global entrepreneurship summit.

Two details tell you this is not another marketing white paper. First, the drafting committee was led by Xinhua News Agency's media-convergence research institutes and the State Key Laboratory of Media Convergence Production Technology and Systems, with more than 30 universities, research institutes, and media groups involved — including Fudan University, Beijing University of Posts and Telecommunications, and Shanghai Jiao Tong University. Second, the standard states plainly that it does not replace advertising law, AI-content-labeling rules, or data-security law; it layers on top of them.

One caveat to keep honest: this is a group standard, not a law with penalties. It carries no administrative fines. But it gives regulators, platforms, and buyers a written answer to the question "what counts as credible?" — and that answer is already being used as a reference in enforcement.

🗂️ Three Libraries, One Rule

The standard's central device is called "three-zone governance." A brand's knowledge base must be split into three libraries and used according to their purpose:

• Fact library — verified facts, each bound to evidence and a source tier.
• Opinion library — commentary, expert views, and analysis, each labeled with its author and time.
• Marketing-expression library — slogans, creative copy, and selling points.

The rule that matters: any factual claim inside marketing copy must match the version in the fact library. Marketing language can no longer stand in for evidence.

This closes the oldest gray-zone shortcut in the AI-visibility playbook — writing an ad that reads like a neutral review so the model treats it as fact. Under the standard, a vendor that packages your selling points as objective test results is running a documented violation, and the brand is the party that wears the trust damage when it is caught.

🏷️ The A-to-D Source Ladder — and Where You Land

The standard defines four source-credibility tiers that decide how much weight a claim carries when an AI model cites it:

• A — information lawfully published by government, judicial, and regulatory bodies; original news from outlets on the official news-source list; effective court rulings; national authoritative databases. Core, priority sources for factual citation.
• B — peer-reviewed papers, independent research reports, association white papers, original reporting by mainstream media, authoritative science platforms. Important evidence sources.
• C — corporate websites, product manuals, commercial reports, user feedback, ordinary media coverage, public web pages. Auxiliary sources; core factual claims require stronger verification.
• D — anonymous sources, untraceable sources, low-quality aggregators, sources with a history of violations. In principle, not to be used for core facts.

The blunt reading for an overseas brand: without a China entity, you are not on the official news-source list, you do not hold an ICP-filed Chinese site or a Baidu Baike entry, and Chinese mainstream media are not quoting you. Your only controlled asset — your corporate website — is a tier-C source by definition. That is a structural ceiling, not a content problem you can write your way out of.

🔍 What "Verifiable" Now Means

The standard converts "credibility" into an auditable record. Every core factual claim must bind nine evidence elements before it is published or fed into a corpus: source tier, source name, original link or file, collection time, last verification time, review owner, validity period, version number, and evidence file number.

Delivery is redefined too. A vendor must keep a full-chain trace — who commissioned, who supplied source material, who generated, optimized, reviewed, confirmed, and published — down to a Trace ID per item. The standard's appendix spells out 20 structured data fields for this trail.

This is the quiet killer of black-box GEO. A vendor who used to hand you a screenshot of "you appeared in an AI answer" now has to hand you a trace that an independent third party could audit. A self-reported screenshot is no longer a deliverable.

📝 What This Changes in Your Vendor Contract

Three clauses rewrite the buying side. First, vendors may not promise or imply "guaranteed number one," "stable top three," "permanent inclusion," "will definitely be cited by the model," "officially recommended partnership," or "exclusive platform cooperation" — absolute-effect claims are now written out as misleading. Second, vendors are graded L1 (basic), L2 (professional), and L3 (integrated governance), a tier intended to become a procurement threshold. Third, an incident-response clock is set: high-risk events need a first response within one hour and initial containment within four hours.

For an overseas brand, these are clauses you can actually enforce. Ask a vendor whether its delivery is independently auditable, whether each core claim carries its nine evidence fields, and how fast it can stop a bad citation. A vendor that cannot answer is out of step with the standard's own terms.

🚀 Climbing from C to B

The standard regulates vendors, but the source ladder is really about your assets. The way out of tier C is to build the sources the standard itself treats as tier B: an ICP-filed Chinese website, a Baidu Baike entry, an institutional Zhihu or Baijiahao presence, and — over time — original coverage by Chinese mainstream media and industry associations.

That is the gap Baidu PPC Pro exists to close for overseas brands. We handle the entity, filing, and account setup you cannot complete without a China presence, then build the Chinese-language visibility assets — Baike, Baijiahao, media relations — that lift a brand from "auxiliary" to "important evidence" in the models' eyes. The standard changed the scoreboard; the game is now won by whoever owns credible sources, not whoever publishes the most.