China’s GEO Standard Exposes What Western Operators Miss

TL;DR: China’s advertising industry released a five-part GEO standard in September 2026, triggered by a national TV scandal showing fabricated AI citations. The standard sets measurement minimums, vendor grading, and a hard line between optimization and manipulation. Western operators spending $10K+ per month on AI visibility services are flying blind without these benchmarks.

A TV Scandal Built the World’s First GEO Rulebook

On March 15, 2026, China Central Television aired its annual consumer protection gala, the most-watched investigative broadcast in the country. The target this year: GEO vendors. An industry insider, filmed on a hidden camera, invented a smart wristband that didn’t exist, then used AI writing tools to produce more than two dozen promotional articles, expert reviews, and user evaluations for the fictional device. When reporters queried mainstream AI models for wristband recommendations, at least two returned the fake product among their top picks. The segment called it “AI poisoning.” The technique required no specialized access, no insider knowledge of any Chinese AI system, and no large budget. It took hours.

What followed was fast by any industry’s standards. By March 16, GEO companies were publishing statements distancing themselves from black-hat operators. China’s market regulator named AI-generated advertising an enforcement priority within the same month. On April 29, the drafting committee held its first full meeting. By September 9 through 14, all five parts of the standard were published on China’s national group standards platform.

The standard was issued by the China Association of Commercial Advertising and organized by its AI Marketing Application Working Committee. It carries a “T/” prefix, meaning it is a voluntary group standard, not law. Its force comes from brands embedding it in contracts and regulators referencing it in enforcement actions. That distinction matters when borrowing from it.

What the Five Parts Actually Cover

The series spans the full lifecycle of a GEO engagement. Part 1 defines GEO as a discipline separate from SEO and AEO, and lays out a four-layer framework. Part 2 grades vendors across six types and five assessment dimensions, including a “one-vote veto” that disqualifies any vendor caught using banned tactics regardless of scores elsewhere. Part 3 specifies pricing models, service units, and the metric formulas that make two vendors’ reports comparable. Part 4 rates source quality for the content corpus and adds stricter rules for regulated industries, including healthcare and finance. Part 5 covers audit trails, contract requirements, and an explicit list of banned tactics.

The list of drafting organizations is as telling as the content. Mininglamp, a Hong Kong-listed marketing technology firm, led Part 3. Content platforms Zhihu, Sohu, Weibo’s operator, and 360 contributed across multiple parts. One name stands out for Western operators: Publicis Group is listed as a drafter of Parts 1 and 5. A major global agency network has already signed its name to these principles in China.

Operators running paid media programs alongside AI visibility strategies should note Part 5’s explicit distinction between legitimate paid promotion and manipulation. The standard protects compliant PR and paid media. It restricts fabrication, undisclosed synthetic content, and citation stuffing.

The Measurement Problem No Western Tool Has Solved

Part 3 opens with an inconvenient truth: generative engines don’t return the same answer twice. Run the same prompt on the same model on the same device, and you will get variation. A single measurement proves almost nothing.

The standard’s answer is a minimum of 30 runs at the finest level you report on. If you report per intent, per engine, per day, that means 30 runs per intent, per engine, every day. The standard backs this with a variance table drawn from a 2025 China GEO industry report. At five runs, a brand that genuinely appears in 50% of answers will show up at 20% or below roughly one report in five, with nothing actually changing in the underlying reality. At 30 runs, uncertainty at 50% true visibility is still plus or minus 7 points on a 68% confidence basis, or roughly 18 points at 95% confidence. These are not comfortable margins for a contract KPI.

The standard defines five core metrics with formulas, so two vendors measuring “visibility” are measuring the same construct. AI visibility rate counts the share of answers that mention or recommend the brand. Share of visibility compares that rate to competitors in the same category. Average position tracks where the brand first appears. Sentiment non-negative rate measures positive or neutral mentions. Source citation rate tracks how often target sources are cited by the model.

Two rules are worth copying immediately. Visibility rate is calculated only on category questions such as “best platform for retail forex trading,” never on branded prompts like “what do you think of Broker X.” Branded prompts almost always surface the brand, so mixing them into the denominator inflates the reported number. And if an AI model is used to score sentiment, the vendor must disclose which model, which prompts, and which rules it applied. Human review must remain in the loop.

If you’re spending real budget on audience targeting and AI search feeds your top-of-funnel, the quality of your visibility measurement directly affects how you interpret lead volume changes. Ask your current vendor how many runs back each number and whether branded prompts are counted.

Where Optimization Ends and Manipulation Begins

This is the question Western SEOs keep debating without resolution. The Chinese standard draws the line in Part 5 with a banned-tactics list and an audit-trail requirement. Fabricating reviews, seeding synthetic citations without disclosure, misrepresenting a product’s attributes in content specifically designed to influence AI training or retrieval, and guaranteeing specific AI ranking positions are all prohibited. The standard does not prohibit publishing high-quality content, earning legitimate media coverage, or running compliant paid campaigns.

The “one-vote veto” in Part 2 means any vendor caught using banned tactics fails the vendor assessment entirely, regardless of how well they score on other dimensions. Brands can require vendors to demonstrate compliance against this framework before signing.

Western operators in regulated verticals have the most immediate exposure here. iGaming acquisition programs, forex lead generation, and legal marketing campaigns all run in categories where regulators are watching AI-generated content closely. A vendor who promises guaranteed AI citation placements in those verticals is either overpromising or using tactics that Part 5 would classify as manipulation. Either way, the operator carries the compliance risk.

What This Means for High-CAC Vertical Operators

If your cost-per-acquisition runs above $200, the quality of your AI visibility data is a budget decision, not just a reporting preference. Forex brokers, iGaming platforms, crypto exchanges, and personal injury law firms all operate in this range. AI citations in those categories influence research-phase traffic that converts at higher rates than paid search cold audiences. Measurement noise at the five-run level can flip a month’s performance narrative without anything real changing.

Three immediate steps apply regardless of vertical. First, run a full marketing audit that includes your current AI visibility reporting methodology. Confirm how many runs back each metric and whether branded prompts are separated. Second, ask any GEO or content vendor to specify which tactics they use against the framework the Chinese standard lays out. If they can’t distinguish between optimization and manipulation in concrete terms, that’s a red flag. Third, build category-question prompt sets for your specific product, then run them at a minimum of 30 times per reporting period before drawing trend conclusions.

Operators running crypto lead programs face an additional layer: AI models are already citing regulatory guidance and risk disclosures prominently in crypto-related queries. Corpus quality, Part 4 of the Chinese standard, determines whether your brand appears alongside credible sources or gets filtered out by the model’s own quality signals.

CDL fleets running driver recruitment campaigns are earlier in the AI visibility curve but not immune. Job-seeker behavior is shifting toward AI-assisted research. Fleets that build structured, credible content now will have a corpus advantage when that shift accelerates.

AI agents used for lead qualification also depend on accurate brand signals in AI models. If the model’s understanding of your brand is built on fabricated or low-quality citations, your agent’s handoff conversations will reflect that misrepresentation. Corpus quality is upstream of every AI touchpoint in the funnel.

What to Take From This Now

The Chinese standard is voluntary and written for a different regulatory environment. It doesn’t map perfectly to Western ad markets. But it answers five questions Western operators should already be forcing their vendors to answer: What is GEO, specifically? How is visibility measured, and with how many samples? How are metrics defined so results are comparable across vendors? What tactics are prohibited? And what does a compliant contract look like?

Western regulators will eventually ask similar questions, with or without a primetime TV scandal to accelerate the timeline. Operators who build measurement discipline and vendor accountability now won’t need to scramble when the rulebook arrives here.

Originally reported by Search Engine Journal, October 2026.