In a segment broadcast in March 2026, an industry expert beinged in front of a hidden electronic camera and designed a smart wristband that didn’t exist.

Software produced an initial batch of more than a loads promotional articles, and he followed up with 11 more: expert testimonials, sector positions, and individual evaluations. When reporters later on asked AI models for smart wristband recommendations, at least 2 listed the fictional gadget among their top picks.

That presentation broadcast on China’s most-watched consumer protection broadcast. It took hours, not months. And nothing about the method depended upon the AI designs being Chinese.

6 months later on, China’s marketing sector responded to with something the Western search market still does not have: a five-part criterion that specifies what generative engine optimization (GEO) is , just how its results ought to be measured, what a GEO contract might guarantee, and where optimization ends and manipulation begins.

I have actually complied with the Chinese search market for several years, and this is the first time I ‘d recommend that Western Search engine optimizations review a Chinese industry requirement carefully. Not because it’s excellent, however because it addresses inquiries our industry keeps arguing about without fixing.

Here’s exactly how it happened, what remains in it, and which concepts deserve loaning.

From Detraction To Standard In 6 Months

To understand why the industry moved this quickly, you require to understand what” 3 · 15 means in China.

Each year on March 15, Globe Customer Rights Day, China Central Tv (CCTV) broadcasts a primetime gala built on undercover examinations into companies that cheat consumers. Past versions have subjected food safety and security offenses, data collecting apps, and phony reviews. Being called on the show can imply governing action, breaking down sales, and public apologies within days.

In 2026, the program took aim at GEO Reporters defined suppliers who promised clients that, for a fee, their products would show up often in mainstream AI answers, also when the material behind those responses was incorrect advertising and marketing. The program called it “AI poisoning.”

What followed moved swiftly:

Why The Market Wrote Its Own Policy

The standard does not claim why it was written in so many words, yet having watched how Chinese sectors reply to 3 · 15 exposés, I see three motives.

Pre-empting regulatory authorities. In China, a 3 · 15 segment is often adhered to by government policies. A market that composes its own common very first reaches shape what “certified” implies before a regulator defines it a lot more strictly.

Conserving the label. Genuine companies had actually invested 2 years selling “GEO” to brand names. After primetime tv linked the term to poisoning AI, they required a clear public line in between themselves and black-hat drivers. As you’ll see below, the standard draws that line clearly, and it protects legitimate PR and paid media as long as it limits adjustment.

Providing buyers a benchmark. Before the criterion, brands had no common method to contrast GEO suppliers, rate their services, or confirm results. Study firm iiMedia, as reported by Yicai , had flagged low standardization and a lack of linked metrics as the young industry’s core problems.

None of these three problems is distinctively Chinese. Western Search engine optimizations deal with the very same mix of overpromising suppliers , inappropriate AI presence metrics , and an undefined line in between optimization and adjustment. We just have not had actually the aired scandal yet.

What The Common Really Covers

The series was issued by the China Organization of Commercial Advertising And Marketing (中国商务广告协会). It’s not to be confused with the separate China Marketing Association. The job was organized by the association’s AI Marketing Application Working Committee. All 5 parts are listed as current on China’s nationwide team criteria platform, which is run by the China National Institute of Standardization.

One factor issues for how you reviewed every little thing below: these are group requirements , marked by the “T/” prefix in their codes. They are voluntary, written by market individuals, and not law. Their pressure comes from brand names composing them into contracts and from regulators using them as a reference.

The list of factors is as fascinating as the content. The platform names loads of drafting companies throughout the 5 parts. Mininglamp, a Hong Kong-listed advertising technology company, led the drafting of Part 3 Content and search systems such as Zhihu, Sohu, Weibo’s driver, and 360 likewise participated.

One name will certainly stick out to Western readers: Publicis Group is noted as a drafter of Parts 1 and 5 A significant Western company network has currently put its name to these principles in China.

See additionally: Baidu Currently Procedures AI Citations: Here’s What SEOs Ought to (And Shouldn’t) Review Into It

5 Concepts Western Search Engine Optimizations Must Obtain

Much of the standard is composed for the Chinese market. Yet 5 of its ideas work anywhere AI responses affect buying choices.

1 Measure AI Exposure With Enough Samples

The trouble every AI exposure device faces is that generative engines do not offer the very same response two times Component 3 says this directly: identical concerns on the exact same device and model can generate various solutions, so a single run confirms extremely little.

The criterion’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, each day, that means 30 runs per intent, per engine, every day. It backs this with a table of the typical deviation of determined visibility at different sample dimensions, drawn from a 2025 industry report by the China Association of Commercial Advertising and marketing’s AI Advertising and marketing Application Working Committee and Mininglamp Modern Technology:

Resource: T/CAACCHINA 003– 2026, Appendix A, pointing out the 2025 China GEO Sector Development Report. Numbers as released. They are one common discrepancy (roughly 68 % self-confidence); a 95 % array has to do with twice as broad. The released values do not adhere to book binomial maths: the 20 % and 80 % columns need to be identical, and the 60 – and 100 -run figures are tighter than theory forecasts. Treat the instructions as strong and the specific values as indicative.

Read the 5 -run row once again. With only 5 runs, ratings can only move in 20 -factor leaps, and a brand name that truly shows up in fifty percent of all answers will certainly land at 20 % or listed below roughly one record in five, with nothing in fact changing. If your monthly AI presence record hinges on a handful of runs per timely, its pattern lines might be noise Also the standard’s minimum isn’t accurate: at 30 runs and 50 % exposure, the 95 % variety is still about ± 18 points.

The standard additionally specifies its core metrics with solutions, so two vendors measuring “visibility” are gauging the same thing:

• AI presence rate: The share of answers that discuss or advise the brand, which can be reported for top- 1 or top- 3 settings.
• Share of visibility: The brand name’s mentions relative to competitors in the category.
• Ordinary position: Where the brand name initially shows up in the response.
• View non-negative rate: The share of states that are positive or neutral.
• Resource citation rate: How often main or target sources are pointed out.

The standard additionally advises that share of visibility and typical placement depend greatly on exactly how the classification and question collection are specified, so they shouldn’t work as standalone contract KPIs.

Two details are worth copying promptly. First, exposure price is determined only on group questions (“finest electric vehicle for families”), never ever on well-known ones (“what do you think of Brand name X”). Branded triggers usually state the brand, so blending them in inflates the number. Second, if an AI model courts view, the supplier needs to divulge which model, which prompts, and which rules it made use of, and maintain human evaluation in the loophole.

What to do: Ask your AI presence device or agency how many runs sit behind each number, and whether branded motivates are counted in presence. If they can not answer, treat their fad lines with care.

2 Put A Checklist Of “Things We Can Not Assurance” In Every Contract

Component 5 bans vendors from promising effects past technological reality. It names two examples: “absolute control of model outcome” and” 100 % screen dominance.”

It after that goes a step better. Agreements must state the scope, timeline, deliverables, and acceptance standards, and they should include a checklist of non-promisable things as an official appendix

Western Search engine optimizations will certainly identify the problem. “We’ll get you advised by ChatGPT” pitches are everywhere, and few of them claim what’s outside the supplier’s control: design updates, retrieval changes, or the basic truth that nobody manages what an LLM says.

What to do: Prior to authorizing any kind of GEO or AI exposure agreement, ask the vendor for a written listing of what they can not assure. A severe vendor will certainly have one. A supplier that stands up to the question has actually informed you something valuable.

3 Structure Pricing So It Doesn’t Reward Gambling

Component 3 explains 5 prices models: project-based, by service units, performance-based, system-based (system licenses or customized growth), and hybrid. The hybrid policies are one of the most useful.

In a hybrid offer, the set part covering web content manufacturing, corpus structure, and tracking should go to least 30 % of the agreement value Efficiency fees are paid in tiers based upon exactly how much targets are satisfied. The agreement should additionally define which engines and setups are measured, the baseline, the targets, and just how model updates or policy adjustments are dealt with.

The reasoning is easy to comply with if you keep in mind the fake wristband. When a vendor makes absolutely nothing unless an AI answer modifications, the cheapest means to transform it is to flooding the web with produced web content. A guaranteed base cost for legitimate job gets rid of some of that pressure.

What to do: If you’re negotiating performance-based GEO prices, settle on the measurement standard and a prepare for model updates before you settle on any type of incentive.

4 Attract A Clear Line In Between Optimization And Control

This is the most transferable idea in the whole criterion.

The standard specifies black-hat GEO through particular strategies: phony material, irregular circulation, corpus poisoning, punctual shot attacks , fake website traffic and evaluations, forged sources, and maliciously disparaging competitors. Agreements should have a stipulation forbiding black-hat GEO, and vendors should decline job targeted at deceptive individuals or averting system policies.

Then comes the important carve-out. Reputable advertising, PUBLIC RELATIONS, media placements, and influencer partnerships are explicitly not black-hat GEO, as long as the material is honest and authorized and absolutely nothing is being adjusted.

That sentence does a great deal of work. It informs brands that gaining coverage, publishing professional material, and collaborating with developers stay reasonable means to affect what AI engines find out. It locates the issue in deception and fabrication, not in the act of attempting to be noticeable.

The standard backs this with a “one-vote veto” in its supplier evaluation. Any company involved in black-hat GEO, corpus poisoning, or comparable adjustment falls short the examination outright, nonetheless strong its various other ratings.

What to do: Write a meaning of black-hat GEO right into your firm contracts. The standard’s checklist of techniques, plus its public relations carve-out, is a great beginning draft.

5 Plan For When AI Obtains Your Brand Name Wrong

Component 5 recommends an error-correction system in an annex. When content a supplier positioned ends up being incorrect, or an AI engine misquotes it, the vendor must start adjustments within 72 hours That implies updating the web content it regulates and aiding the customer upgrade its internet site, knowledge bases, and other channels.

The very same annex calls for an emergency situation feedback. If a corpus error creates a significant case, such as a wrong rate that relocates a supply or sets off consumer grievances, associated GEO task need to stop.

Part 4 additionally establishes stricter sourcing rules for sensitive industries such as medical care and finance. There, only whitelisted resources matter, and confidential posts and unproven customer content are dismissed. Western SEOs will certainly identify this as a relative of Google’s “Your Cash or Your Life” (YMYL) reasoning.

What to do: Treat AI misinformation concerning your brand name like any various other credibility incident. Determine in advance who monitors AI solutions, that fixes the source web content, and just how quick.

See additionally: Brand Protection In AI Look: How To Audit And Defend Your Brand’s Identity

Where To Remain Unconvinced

A criterion is only just as good as the incentives of individuals who wrote it. Before you take on any one of the ideas over, maintain these restrictions in mind.

It was created by the market it regulates. The drafters are mainly GEO vendors, companies, and marketing modern technology companies. Mininglamp led the drafting of Part 3, which covers rates and measurement, and it co-authored the 2025 market record with the board that organized the job. That does not make the material wrong, yet regulations composed by vendors tend to shield the market as high as the customer.

The 30 -run regulation originates from inside the room. The basic mentions a 2025 market record by the very same board and Mininglamp as the basis for its sample-size table. The underlying principle is fundamental tasting stats, so I would certainly expect independent tests to verify the instructions. Still, deal with the exact numbers generally of thumb, not resolved science.

It’s volunteer. Absolutely nothing compels a supplier to follow a group criterion. Its actual power will originate from brand names making compliance an agreement condition, and from regulatory authorities treating it as a benchmark of good method. We’ll only understand in a year or 2 whether either happens.

Some of it is old search engine optimization with brand-new labels Part 1’s four-layer structure (technological structure, content high quality, platform adaptation, brand name mind share) maps carefully to technological search engine optimization, E-E-A-T, and brand name positioning. The standard even points out Google’s E-E-A-T by name, with a note that it hasn’t been verified for generative engines. Western viewers won’t find much brand-new there, and that’s penalty. The worth depends on the measurement, contract, and conformity guidelines.

It was constructed for a various ecosystem. The requirement 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 setting doesn’t.

Does The West Requirement Its Own 3 · 15 Minute?

The phony wristband dealt with Chinese AI models. I see no reason to believe the very same technique would stop working on Western ones. Retrieval-based solutions anywhere compensate volume, rep, and sources that look authoritative.

The difference is presence. China had a primetime demo, a regulatory authority’s action, and a sector that needed to fix its online reputation quickly. In the West, comparable manipulation can take place quietly , and nobody has yet had a solid enough reason to create the regulations down.

If a Western version ever before comes, that would certainly write it? Industry associations are one candidate. The AI systems are an additional, since they have one of the most to lose if users stop trusting solutions. Especially, China’s effort consisted of content platforms like Zhihu and Weibo, the type of resources AI engines attract from. Could Reddit, evaluation platforms, and significant publishers sit at a comparable table?

Or the push can come from purchasers. Brand names that collectively demand example dimensions, non-promise lists, and black-hat provisions in their agreements would certainly create a requirement in method, without any person publishing one.

You do not need to await any one of that. Every concept in this post can go into your next dimension setup or company contract:

• Run each tracked prompt enough times to rely on the number, and maintain branded prompts out of presence prices.
• Ask vendors for a composed list of what they can not assure.
• Keep a fixed cost for legitimate operate in performance-based bargains.
• Specify black-hat GEO in writing, with a specific carve-out for sincere PR and web content.
• Have a prepare for fixing AI mistakes about your brand name within days, not months.

China needed a nationwide TV scandal to get there. The rest people can merely review their rulebook.

A lot more Resources:

Included Image: Pawel Michalowski/Shutterstock