# China Wrote The First GEO Rulebook After A TV Scandal – Western SEOs Should Take Notes > Source: > Published: 2026-10-21 12:00:37+00:00 In a segment aired in March 2026, an industry insider sat in front of a hidden camera and invented a smart wristband that didn’t exist. [Software generated](https://www.chinadaily.com.cn/a/202603/15/WS69b6be6aa310d6866eb3de9e.html) a first batch of more than a dozen promotional articles, and he followed up with 11 more: expert reviews, industry rankings, and user evaluations. When reporters later asked AI models for smart wristband recommendations, at least two listed the fictional device among their top picks. That demonstration aired on China’s most-watched consumer protection broadcast. It took hours, not months. And nothing about the technique depended on the AI models being Chinese. Six months later, China’s advertising industry answered with something the Western search industry still doesn’t have: a five-part standard that [defines what generative engine optimization (GEO) is](https://www.searchenginejournal.com/a-little-clarity-on-seo-geo-and-aeo/565522/), how its results should be measured, what a GEO contract may promise, and where optimization ends and manipulation begins. I’ve followed the Chinese search market for years, and this is the first time I’d recommend that Western SEOs read a Chinese industry standard closely. Not because it’s perfect, but because it answers questions our industry keeps arguing about without resolving. Here’s how it came about, what’s in it, and which ideas are worth borrowing. ## From Scandal To Standard In 6 Months To understand why the industry moved this fast, you need to know what “3·15” means in China. Every year on March 15, World Consumer Rights Day, China Central Television (CCTV) airs a primetime gala built on undercover investigations into companies that cheat consumers. Past editions have exposed food safety violations, data harvesting apps, and fake reviews. Being named on the show can mean regulatory action, collapsing sales, and public apologies within days. In 2026, [the show took aim at GEO](https://www.globaltimes.cn/page/202603/1357055.shtml). Reporters described vendors who promised clients that, for a fee, their products would appear frequently in mainstream AI answers, even when the content behind those answers was false advertising. The program called it “AI poisoning.” What followed moved quickly: | Date (2026) | What happened | |---|---| | March 15 | CCTV’s 3·15 Gala exposes fabricated content, AI data poisoning, and ranking manipulation by GEO vendors. | | March 16 | Several GEO companies publish statements distancing themselves from black-hat operators. | | March | China’s market regulator (SAMR) names citation-based and AI-generated advertising as enforcement priorities. | | April 29 | The standard’s drafting committee (caacchina.org) holds its first full meeting in Beijing. | | September 9–14 | All five parts of the GEO standard are issued and published on China’s national group standards platform. | ## Why The Industry Wrote Its Own Rules The standard doesn’t say why it was written in so many words, but having watched how Chinese industries respond to 3·15 exposés, I see three motives. **Pre-empting regulators.** In China, a 3·15 segment is often followed by government rules. An industry that writes its own standard first gets to shape what “compliant” means before a regulator defines it more strictly. **Saving the label.** Legitimate agencies had spent two years selling “GEO” to brands. After primetime television linked the term to poisoning AI, they needed a clear public line between themselves and black-hat operators. As you’ll see below, the standard draws that line explicitly, and it protects legitimate PR and paid media as much as it restricts manipulation. **Giving buyers a yardstick.** Before the standard, brands had no shared way to compare GEO vendors, price their services, or verify results. Research firm iiMedia, [as reported by Yicai](https://www.yicaiglobal.com/news/chinas-annual-cctv-consumer-rights-show-uncovers-ai-ad-tricks-that-deceive-customers), had flagged low standardization and a lack of unified metrics as the young industry’s core problems. None of these three problems is uniquely Chinese. Western SEOs face the same mix of [overpromising vendors](https://www.searchenginejournal.com/googles-updated-guidance-urges-ftc-complaints-against-shady-seos/578151/), incompatible [AI visibility metrics](https://www.searchenginejournal.com/brands-are-tracking-ai-visibility-but-are-they-measuring-the-right-things/587888/), and an undefined line between optimization and manipulation. We just haven’t had the televised scandal yet. ## What The Standard Actually Covers The series was issued by the China Association of Commercial Advertising (中国商务广告协会). It’s not to be confused with the separate China Advertising Association. The work was organized by the association’s AI Marketing Application Working Committee. All five parts are listed as current on China’s national group standards platform, which is operated by the China National Institute of Standardization. One point matters for how you read everything below: these are *group standards*, marked by the “T/” prefix in their codes. They are voluntary, written by industry participants, and not law. Their force comes from brands writing them into contracts and from regulators using them as a reference. | Code | Official English title | What it does | |---|---|---| | [T/CAACCHINA](https://www.ttbz.org.cn/standardDetail/0047401fbeff49148ed25291fc75c245.html) 001—2026 | Part 1: General, terminology and definitions | Defines GEO, separates it from SEO and AEO, and sets out a four-layer framework. | | [T/CAACCHINA](https://www.ttbz.org.cn/standardDetail/21bfb206cb674888902cb0bef643fbe2.html) 002—2026 | Part 2: Service provider assessment specification | Six vendor types, five assessment dimensions, A–D grades, and a “one-vote veto.” | | [T/CAACCHINA](https://www.ttbz.org.cn/standardDetail/6f9a3b84c9a54cf48600f395b115e7b5.html) 003—2026 | Part 3: Pricing, evaluation and measurement specification | Service units, pricing models, and metric formulas with sampling rules. | | [T/CAACCHINA](https://www.ttbz.org.cn/standardDetail/42f415cbf9b243908de5fc3fddc671db.html) 004—2026 | Part 4: Trusted corpus specification | Source quality ratings and stricter rules for regulated industries, and correction duties. | | [T/CAACCHINA](https://www.ttbz.org.cn/standardDetail/a7b9efdac5d8402dbfee09bb0fec0704.html) 005—2026 | Part 5: Service process compliance and security requirements | Audit trails, banned tactics, and contract requirements. | The list of contributors is as interesting as the content. The platform names dozens of drafting organizations across the five parts. Mininglamp, a Hong Kong-listed marketing technology company, led the drafting of Part 3. Content and search platforms such as Zhihu, Sohu, Weibo’s operator, and 360 also took part. One name will stand out to Western readers: *Publicis Group* is listed as a drafter of Parts 1 and 5. A major Western agency network has already put its name to these principles in China. *See also: [Baidu Now Measures AI Citations: Here’s What SEOs Should (And Shouldn’t) Read Into It](https://www.searchenginejournal.com/baidu-now-measures-ai-citations-heres-what-seos-should-and-shouldnt-read-into-it/590276/)* ## 5 Ideas Western SEOs Should Borrow Much of the standard is written for the Chinese market. But five of its ideas work anywhere AI answers influence buying decisions. ### 1. Measure AI Visibility With Enough Samples The problem every AI visibility tool faces is that [generative engines don’t give the same answer twice](https://www.searchenginejournal.com/how-to-measure-ai-search-visibility/579893/). Part 3 says this directly: identical questions on the same device and model can produce different answers, so a single run proves very little. The standard’s answer is a minimum of 30 runs at the finest level you report on. Its own example: if you report per intent, per engine, per day, that means 30 runs per intent, per engine, every day. It backs this with a table of the standard deviation of measured visibility at different sample sizes, taken from a 2025 industry report by the China Association of Commercial Advertising’s AI Marketing Application Working Committee and Mininglamp Technology: | Runs per question | True visibility 20% | True visibility 50% | True visibility 80% | |---|---|---|---| | 5 | ±18 points | ±24 points | ±15 points | | 10 | ±13 points | ±16 points | ±10 points | | 30 | ±7 points | ±8 points | ±6 points | | 60 | ±4 points | ±5 points | ±4 points | | 100 | ±2 points | ±2 points | ±2 points | Source: T/CAACCHINA 003—2026, Appendix A, citing the 2025 China GEO Industry Development Report. Figures as published. They are one standard deviation (roughly 68% confidence); a 95% range is about twice as wide. The published values don’t follow textbook binomial maths: the 20% and 80% columns should be identical, and the 60- and 100-run figures are tighter than theory predicts. Treat the direction as solid and the exact values as indicative. Read the 5-run row again. With only five runs, scores can only move in 20-point jumps, and a brand that truly appears in half of all answers will land at 20% or below roughly one report in five, with nothing actually changing. If your monthly AI visibility report rests on a handful of runs per prompt, its [trend lines may be noise](https://www.searchenginejournal.com/ai-visibility-rankings-arent-stable-new-research-shows-its-mostly-statistical-noise/581905/). Even the standard’s minimum isn’t precise: at 30 runs and 50% visibility, the 95% range is still about ±18 points. The standard also defines its core metrics with formulas, so two vendors measuring “visibility” are measuring the same thing: - **AI visibility rate:** The share of answers that mention or recommend the brand, which can be reported for top-1 or top-3 positions. - **Share of visibility:** The brand’s mentions relative to competitors in the category. - **Average position:** Where the brand first appears in the answer. - **Sentiment non-negative rate:** The share of mentions that are positive or neutral. - **Source citation rate:** How often official or target sources are cited. The standard also warns that share of visibility and average position depend heavily on how the category and question set are defined, so they shouldn’t serve as standalone contract KPIs. Two details are worth copying immediately. First, visibility rate is calculated **only on category questions** (“best electric car for families”), never on branded ones (“what do you think of Brand X”). Branded prompts almost always mention the brand, so mixing them in inflates the number. Second, if an AI model judges sentiment, the vendor must disclose which model, which prompts, and which rules it used, and keep human review in the loop. **What to do:** Ask your [AI visibility tool or agency](https://www.searchenginejournal.com/we-need-to-change-our-approach-to-ai-prompt-tracking/579646/) how many runs sit behind each number, and whether branded prompts are counted in visibility. If they can’t answer, treat their trend lines with caution. ### 2. Put A List Of “Things We Can’t Promise” In Every Contract Part 5 bans vendors from promising effects beyond technical reality. It names two examples: “absolute control of model output” and “100% screen dominance.” It then goes a step further. Contracts must state the scope, timeline, deliverables, and acceptance criteria, and they must include a list of non-promisable items **as a formal appendix**. Western SEOs will recognize the problem. “We’ll get you recommended by ChatGPT” pitches are everywhere, and few of them say what’s outside the vendor’s control: model updates, retrieval changes, or the simple fact that no one controls what an LLM says. **What to do:** Before signing any GEO or AI visibility contract, ask the vendor for a written list of what they cannot promise. A serious vendor will have one. A vendor who resists the question has told you something useful. ### 3. Structure Pricing So It Doesn’t Reward Gambling Part 3 describes five pricing models: project-based, by service units, performance-based, system-based (platform licenses or custom development), and hybrid. The hybrid rules are the most practical. In a hybrid deal, the fixed portion covering content production, corpus building, and monitoring **should be at least 30% of the contract value**. Performance fees are paid in tiers based on how far targets are met. The contract must also define which engines and settings are measured, the baseline, the targets, and how model updates or rule changes are handled. The logic is easy to follow if you remember the fake wristband. When a vendor earns nothing unless an AI answer changes, the cheapest way to change it is to flood the web with fabricated content. A guaranteed base fee for legitimate work removes some of that pressure. **What to do:** If you’re negotiating performance-based GEO pricing, agree on the measurement baseline and a plan for model updates before you agree on any bonus. ### 4. Draw A Clear Line Between Optimization And Manipulation This is the most transferable idea in the whole standard. The standard defines black-hat GEO through specific tactics: fake content, abnormal distribution, corpus poisoning, [prompt injection attacks](https://www.searchenginejournal.com/prompt-injection-just-proved-something-seo-has-known-for-25-years/586405/), fake traffic and reviews, forged sources, and maliciously disparaging competitors. Contracts must contain a clause prohibiting black-hat GEO, and vendors should refuse work aimed at misleading users or evading platform rules. Then comes the crucial carve-out. Legitimate advertising, PR, media placements, and influencer collaborations are explicitly **not** black-hat GEO, as long as the content is truthful and lawful and nothing is being manipulated. That sentence does a lot of work. It tells brands that earning coverage, publishing expert content, and working with creators remain fair ways to influence what AI engines learn. It locates the problem in deception and fabrication, not in the act of trying to be visible. The standard backs this with a “one-vote veto” in its vendor assessment. Any provider involved in black-hat GEO, corpus poisoning, or similar manipulation fails the evaluation outright, however strong its other scores. **What to do:** Write a definition of black-hat GEO into your agency agreements. The standard’s list of tactics, plus its PR carve-out, is a good starting draft. ### 5. Plan For When AI Gets Your Brand Wrong Part 5 recommends an error-correction mechanism in an annex. When content a vendor placed turns out to be wrong, or an AI engine misquotes it, the vendor should **start corrections within 72 hours**. That means updating the content it controls and helping the client update its website, knowledge bases, and other channels. The same annex calls for an emergency response. If a corpus error causes a major incident, such as a wrong price that moves a stock or triggers consumer complaints, related GEO activity should pause. Part 4 also sets stricter sourcing rules for sensitive industries such as healthcare and finance. There, only whitelisted sources count, and anonymous posts and unverified user content are ruled out. Western SEOs will recognize this as a cousin of Google’s [“Your Money or Your Life” (YMYL)](https://www.searchenginejournal.com/how-to-create-health-ymyl-content-that-performs-in-ai-search/584431/) logic. **What to do:** Treat [AI misinformation](https://www.searchenginejournal.com/biggest-ai-search-risk-is-conflicting-information/586904/) about your brand like any other reputation incident. Decide in advance who monitors AI answers, who fixes the source content, and how fast. *See also: [Brand Protection In AI Search: How To Audit And Defend Your Brand’s Identity](https://www.searchenginejournal.com/brand-protection-in-ai-search-how-to-audit-and-defend-your-brands-identity/587581/)* ## Where To Stay Skeptical A standard is only as good as the incentives of the people who wrote it. Before you adopt any of the ideas above, keep these limits in mind. **It was written by the industry it regulates.** The drafters are mostly GEO vendors, agencies, and marketing technology companies. Mininglamp led the drafting of Part 3, which covers pricing and measurement, and it co-authored the 2025 industry report with the committee that organized the work. That doesn’t make the content wrong, but rules written by sellers tend to protect the market as much as the buyer. **The 30-run rule comes from inside the room.** The standard cites a 2025 industry report by the same committee and Mininglamp as the basis for its sample-size table. The underlying principle is basic sampling statistics, so I’d expect independent tests to confirm the direction. Still, treat the exact numbers as a rule of thumb, not settled science. **It’s voluntary.** Nothing forces a vendor to follow a group standard. Its real power will come from brands making compliance a contract condition, and from regulators treating it as a benchmark of good practice. We’ll only know in a year or two whether either happens. **[Some of it is old SEO with new labels](https://www.searchenginejournal.com/googles-new-ai-search-guide-calls-aeo-and-geo-still-seo/575026/).** Part 1’s four-layer framework (technical foundation, content quality, platform adaptation, brand mind share) maps closely to technical SEO, E-E-A-T, and brand positioning. The standard even cites Google’s E-E-A-T by name, with a note that it hasn’t been validated for generative engines. Western readers won’t find much new there, and that’s fine. The value lies in the measurement, contract, and compliance rules. **It was built for a different ecosystem.** The standard was written with DeepSeek, Doubao, Kimi, and other Chinese engines in mind, in a market where a state broadcaster’s exposé can trigger enforcement within days. The ideas transfer to ChatGPT, Gemini, Perplexity, and AI Overviews. The enforcement environment doesn’t. ## Does The West Need Its Own 3·15 Moment? The fake wristband worked on Chinese AI models. I see no reason to believe the same technique would fail on Western ones. Retrieval-based answers everywhere reward volume, repetition, and sources that look authoritative. The difference is visibility. China had a primetime demonstration, a regulator’s response, and an industry that needed to repair its reputation fast. In the West, [similar manipulation can happen quietly](https://www.searchenginejournal.com/googles-spam-update-now-reaches-ai-answers-enforcement-is-hard/580535/), and nobody has yet had a strong enough reason to write the rules down. If a Western version ever comes, who would write it? Industry associations are one candidate. The AI platforms are another, since they have the most to lose if users stop trusting answers. Notably, China’s effort included content platforms like Zhihu and Weibo, the kinds of sources AI engines draw from. Could Reddit, review platforms, and major publishers sit at a similar table? Or the push could come from buyers. Brands that collectively demand sample sizes, non-promise lists, and black-hat clauses in their contracts would create a standard in practice, without anyone publishing one. You don’t have to wait for any of that. Every idea in this article can go into your next measurement setup or agency contract: - Run each tracked prompt enough times to trust the number, and keep branded prompts out of visibility rates. - Ask vendors for a written list of what they cannot promise. - Keep a fixed fee for legitimate work in performance-based deals. - Define black-hat GEO in writing, with an explicit carve-out for honest PR and content. - Have a plan for correcting AI errors about your brand within days, not months. China needed a national TV scandal to get there. The rest of us can simply read their rulebook. **More Resources:** - [Why International SEO Needs A Global Knowledge Integrity Strategy](https://www.searchenginejournal.com/why-international-seo-needs-a-global-knowledge-integrity-strategy/577880/) - [Google’s New Guidance Claims Authority Over SEO, Tools, And AEO/GEO](https://www.searchenginejournal.com/googles-new-guidance-claims-authority-over-seo-tools-and-aeo-geo/578162/) - [AI Poisoning: Black Hat SEO Is Back](https://www.searchenginejournal.com/ai-poisoning-black-hat-seo-is-back/561217/) *Featured Image: Pawel Michalowski/Shutterstock*