Key takeaway: China’s commercial advertising industry issued a five-part group standard for generative engine optimization between September 9 and 14, 202.
Key Takeaway
China’s commercial advertising industry issued a five-part group standard for generative engine optimization between September 9 and 14, 2026. The framework covers how brands and vendors measure AI search visibility, assess service providers, set prices, manage trusted source material, and document compliant processes. It was organized by the AI Marketing Application Working Committee of the China Association of Commercial Advertising and published on China’s national group standards platform. The rules are voluntary industry standards, not legislation; their practical influence comes from contracts and possible use as a regulatory reference.
How the standard measures AI search visibility
The framework separates GEO from SEO and AEO and defines metrics including AI visibility rate, share of visibility, average position, non-negative sentiment rate, and source citation rate. It says measurement should use enough repeated runs to account for variation in generative answers. At the finest reporting level, it sets a minimum of 30 runs. Category prompts such as “the best electric car for families” should be used for visibility-rate calculations, while branded prompts should not be mixed into that metric because they already name the brand and can inflate the result.
The standard also calls for disclosure of the engines, settings, baselines, targets, and treatment of model or rules updates used in a measurement program. Share of visibility and average position depend heavily on how a category and question set are defined, so the document says they should not be used as standalone contract KPIs.
What GEO contracts must address
The five parts cover provider assessment, pricing and measurement, trusted corpora, and service-process compliance and security. Contracts should specify scope, timing, deliverables, and acceptance criteria, and include a formal list of outcomes that cannot be promised. The standard gives absolute control of model output and 100% screen dominance as examples of impermissible promises.
For hybrid pricing, it says the fixed portion covering content production, corpus building, and monitoring should account for at least 30% of contract value. Performance fees should be tiered according to target achievement. The framework also identifies fabricated content, abnormal distribution, corpus poisoning, and ranking manipulation as black-hat GEO tactics. Brands buying AI marketing or AI search services can use these provisions to ask vendors how results are sampled, which engines are included, how updates are handled, and which outcomes remain outside the vendor’s control.
Method and Limitations
Readers can verify the five documents through China’s national group standards platform by checking their codes, titles, publication records, and scope statements. Buyers can also request the prompt set, run count, model settings, and human-review rules behind each reported visibility metric. The public description identifies these documents as voluntary group standards written by industry participants, so they should not be treated as laws or as proof that any particular vendor can control an AI answer. Results also remain dependent on prompt selection, category definitions, models, devices, and sampling methods; comparisons that omit those conditions may be misleading.