China’s first national standard focused on the relationship between human and intelligent customer service took effect on September 1. The official National Standard Information Public Service Platform lists GB/T 47746-2026 as the national standard governing the coordination of live and intelligent customer-support channels, released on May 25 and effective from September 1.
The standard arrives as automated customer support becomes routine across Chinese platforms and service businesses. Reporting by Xinhua describes frequent consumer complaints about AI answers that miss the question, difficult routes to a human agent, and conflicting commitments between automated and live channels. The new standard does not ban AI customer service. It sets an operating framework for when automation should handle routine matters and when a person or a defined process must take control.
The Standard Defines a More Usable Human Handoff
Xinhua says intelligent customer service should take priority for standardized questions, while human agents should take priority for complex issues and matters involving safety. When a customer moves from an automated system to a person, the standard calls for a convenient entry point and a smooth connection. It also says customer information should transfer across the handoff so the person does not ask the user to repeat the same account.
That requirement goes to the heart of many complaints about automated support. A chatbot may collect an order number, a description of a fault, and a preferred resolution, only for the human agent to begin the conversation again. The standard treats continuity as part of collaboration, not merely a feature of user-interface design. It therefore raises the bar for companies that portray an AI system and a live support channel as parts of one service.
The policy follows a closely related consumer-protection debate. EastFrontier’s August 26 coverage of the China Consumer Association’s AI-service liability guidance examined responsibility for what automated systems say to users. GB/T 47746-2026 is a different instrument. It focuses on the operating relationship between human and intelligent service, including how cases move across that boundary.
The standard’s public status needs to be described precisely. The “GB/T” designation identifies it as a recommended national standard, not a compulsory national standard. It can shape service quality expectations, provide a benchmark for regulators or dispute resolution, and influence corporate design choices. It does not by itself make every AI support failure an automatic legal violation.
Prices, Refunds, and Contracts Require More Than a Chatbot Reply
The new framework draws a clear line around decisions with financial or contractual consequences. Xinhua says ordinary questions and product introductions may be answered automatically, but AI-generated responses should carry a prominent label and risk notice. Where a conversation concerns prices, discounts, refunds, compensation, or contract changes, an AI system may explain rules, collect materials, or generate a work order. Final confirmation, however, should come from a human customer-service agent or a standardized process.
This is an operational rule with broad relevance. E-commerce marketplaces, telecommunications providers, financial platforms, transport services, and utility companies all use automated channels for matters that can affect a customer’s money or rights. The standard tells companies to separate information and intake from the final confirmation step in these higher-stakes cases.
It also addresses an ambiguity that can frustrate users. A company may place an AI system in its official service channel, permit it to look up orders or create cases, and then later reject the system’s answer as nonbinding. The standard does not resolve every possible legal dispute, but its approach pushes companies toward clearer disclosures, defined authority boundaries, and a reliable path for human review.
China has increasingly treated AI standards as part of implementation policy rather than a purely technical back-office issue. EastFrontier’s August 27 examination of China’s nearly 200 AI standards highlighted the policy work required to turn the country’s AI Plus agenda into operational practice. Customer service is a useful test case because it is where consumers experience AI systems directly and at high volume.
Recommended Rules Can Still Change Corporate Behavior
A recommended standard can influence conduct even without a direct statutory penalty. Xinhua reports that the document offers regulators and courts a possible reference point for service-quality assessment and disputes. It also notes that existing laws already require businesses to maintain convenient and effective complaint-handling mechanisms. A company that creates repeated obstacles to human support may face questions under those broader obligations.
For businesses, the immediate task is to map what their AI systems can do. Automated tools should have explicit limits on the data they can access, the actions they can take, and the commitments they can present to customers. Human escalation should be a designed workflow, not a hidden fallback. Context transfer must work in practice, because the standard specifically identifies repeated questioning as a problem to avoid.
For customers, the standard should improve the vocabulary for assessing service quality. A user who is trapped in a loop or forced to repeat information can point to a national framework that expects convenient transfer and continuity. Whether that produces better outcomes will depend on implementation, enforcement, and the willingness of firms to treat resolution quality as more important than reducing the number of cases reaching people.
China’s September 1 standard does not settle the debate over automated accountability. It does, however, articulate a concrete model: AI can handle routine work, but businesses should preserve human judgment and procedural safeguards when a customer’s safety, money, or contractual position is at stake.