China Cracks Down on AI Customer Service ‘Invisible Filters’

China has put its first national standard governing the coordination between human and AI customer service into effect, a move aimed at curbing widespread consumer complaints about being trapped in endless chatbot loops and unable to reach a human agent. The standard, GB/T 47746-2026, “Customer Contact Service: Requirements for Collaboration Between Human and Intelligent Customer Service,” took effect on September 1, 2026, according to People’s Daily, which published an in-depth feature on the measure on September 11.

Issued by the State Administration for Market Regulation (SAMR) and the Standardization Administration of China (SAC), the recommended national standard applies to a rapidly expanding field: AI customer service has been embedded across e-commerce, telecommunications, finance, travel, and home-appliance sectors. While it has boosted response efficiency and cut operating costs, the technology has often become what regulators and consumer advocates describe as a “roadblock” rather than a service aid.

A Growing Consumer Backlash

The scale of the problem is significant. According to the China Consumers Association (CCA), consumer associations nationwide handled 986,000 complaints in the first half of 2026 alone, with after-sales service accounting for the largest share at 26.79%. False promises by AI agents, difficulty accessing human representatives, and inaccurate AI-generated content have emerged as new complaint hotspots. A search for “AI customer service” on a third-party consumer complaint platform returns more than 4,800 related complaints.

The frustration is captured in the everyday experiences of consumers. Ms. Wu from Xuzhou, Jiangsu province, repeatedly typed “transfer to human” only to have the AI keep replying with standardized menus — it took her over an hour to finally reach a person who resolved her return request in a few sentences. “It’s too much trouble,” she told People’s Daily. In another case, a family in Guangdong province was locked out of their home in June 2026 when their smart door lock malfunctioned, and the after-sales line could not transfer them to a human agent — leaving them, including a six-month-old baby, waiting outside for 50 minutes before hiring a locksmith.

“On some platforms, AI customer service is no longer purely a technical tool but has been designed into an ‘invisible filter net,’” said Guo Yaozong, deputy director of the CCA’s complaint department.

The Economics Behind the Friction

The drive to cut costs lies at the heart of the trend. A human customer service agent costs a company at least roughly RMB 3,000 per month when wages, social insurance, training, and workstations are factored in, while a basic AI customer service system rents for as little as RMB 99 per month and can handle the workload of dozens of human agents — a roughly 30-fold cost gap.

“The customer service department is defined in many companies as a ‘cost center,’ and the primary KPI is cutting costs,” said Chen Yinjiang, deputy secretary-general of the Consumer Rights Protection Law Research Institute at the China Law Society. He noted that some companies now measure “AI interception rate” rather than resolution rate, with some “obstacle designs” able to intercept up to 70% of human-transfer requests.

Su Haopeng, a professor at the University of International Business and Economics School of Law, argued that this is a misguided approach. “On the surface, customer service costs go down, but the problem is not solved — it is merely postponed and concealed, transforming into repeated inquiries, public-opinion fallout, lost returns, and brand damage,” he said. “These hidden costs far exceed the labor costs saved.”

What the New Standard Requires

According to Wu Yansong, director of the Caibo Smart Governance Research Institute and a participant in drafting the standard, the measure sets out four core requirements:

• Prominent entry: Human-agent entry points must be clearly and prominently displayed, and hiding them behind layers of menus is prohibited.
• Clear division of labor: Complex disputes, financial disputes, personal or property safety risks, and any situation where a user explicitly requests a transfer must be promptly handled by a human. Key matters such as pricing, refunds, and compensation must be finally confirmed by a human agent.
• Smooth switching: When switching between human and AI service, the system must fully transfer the interaction record, and consumers must not be asked to repeat information they have already provided.
• Clear responsibility: Companies are responsible for the content of AI customer service replies and cannot shirk responsibility on the grounds that content was “auto-generated by algorithm.”

A separate report by Xinhua, citing legal experts, noted that the standard also requires AI customer service to offer multiple switching methods — keyword triggers, menu selection, key presses, and voice commands — and to automatically transfer to a human when interactions fail, negative emotion is detected, or safety-related scenarios arise.

A ‘Toothless’ Standard? Not Quite

Because it is a recommended, rather than mandatory, standard, the measure carries no direct penalties on its own. But legal experts and regulators argue it is far from toothless.

“This national standard is a recommended standard, but it is by no means ‘toothless,’” said Liao Huaixue, legal counsel for the CCA. It will serve as an important reference for regulators’ administrative inspections and service-quality evaluations, and non-compliant firms may be required to rectify. It will also provide a basis for consumer rights advocacy and judicial adjudication in disputes.

Wu echoed this, noting that market regulators can use the standard as an important reference when handling cases that infringe on consumer rights under the Consumer Rights Protection Law, and that the CCA and courts can likewise treat it as a benchmark for industry practice. Companies that fall short may face reputational damage and customer loss, creating a market-driven incentive to comply.

The standard joins a broader policy “combination punch.” The Interim Measures for the Administration of Anthropomorphic AI Interactive Services took effect on July 15, 2026, requiring providers to fulfill safety responsibilities and deploy content management measures. The Consumer Rights Protection Law Implementation Regulations and the E-Commerce Law also require operators to establish convenient complaint-handling mechanisms. The CCA has issued a special statement urging operators to strengthen review and management, and consumer associations in Zhejiang, Guangdong, and Beijing have held discussions with companies.

Enforcement Reality and What’s Next

The gap between intention and practice was evident from day one. On September 1, a Henan Television reporter tested mainstream social, shopping, and travel apps and found that most major platforms did not offer a direct “human service” entry — users had to instruct the AI to transfer, and some required repeated requests before being allowed through. Some platforms’ bots remained in “offline-mode,” replying only with fixed templates.

Still, some leading companies have begun adjusting. Some have moved the “transfer to human” button to app home pages, and others are exploring deeper “AI + human” collaboration, achieving lower after-sales costs alongside improved user satisfaction. On September 9, JD.com announced an upgrade of its app to version 16.0 featuring an AI shopping assistant named “Dongdong,” underscoring both the promise and the scrutiny facing the sector.

The legal foundation is also firming up. Under China’s Consumer Rights Protection Law, operators have a duty to give truthful, clear answers to consumer inquiries — an obligation unaffected by whether the service carrier is human or AI. Legal experts have said that companies invoking “AI answers do not represent the company’s position” to evade responsibility is legally untenable.

“Technology empowers consumption, but service must have warmth,” Chen said, arguing that the key is to push companies from a “cost-control orientation” to a “consumer-experience orientation.”

For consumers, Guo of the CCA offered practical advice: decisions involving pricing, contracts, and after-sales terms should not rest solely on AI-generated content, and those who encounter difficulty reaching a human should preserve evidence and file complaints through the 12315 hotline or local consumer associations.

As the standard takes hold, the question is whether it can close the gap between what companies promise and what their chatbots actually deliver — and whether “one-click transfer to human” becomes the norm rather than the exception.