China’s consumer watchdog is drawing a practical line around AI use in commerce: a generated answer is not a license to mislead a customer, and a company cannot erase its responsibilities by calling an error “algorithmic.” The China Consumer Association issued new consumer guidance on August 25 that separates public generative-AI services from AI customer-service systems used by merchants, while warning consumers to verify important information through official channels.
The distinction is straightforward but important. A general-purpose AI assistant may help someone understand a product, contract, or service problem, but it is not automatically speaking for the business involved. An AI customer-service tool embedded in a merchant’s own sales and support process is different. According to Securities Times, the association said businesses cannot simply avoid responsibility for information about pricing, fees, promotions, contracts, or after-sales service by saying the reply was produced automatically.
The guidance is not a new law. It does not create a standalone licensing regime or prescribe a single technical solution. It is a consumer-protection intervention that tells companies where existing responsibilities should continue to apply as AI systems enter more customer interactions. CCTV News reported that the association advised consumers not to make important choices about prices, contract terms, after-sales matters, or legal liability solely on the basis of AI-generated content.
Two Different Kinds of AI Consumer Risk
The association’s framework starts by separating two situations that are often blurred together. The first is a public generative-AI service. These systems answer questions based on model behavior, available information, and a user’s prompt. The association describes their outputs as reference material rather than an official commitment by a platform, merchant, or other party. A user who asks a chatbot about a refund, a financial product, or a legal dispute therefore needs to check the answer against an official policy, contract, regulator, or other authoritative source.
That warning reflects a simple fact about generative systems: fluent language can create a false sense of certainty. An answer may sound precise even when it leaves out an exception, misunderstands the question, or invents a detail. In consumer settings, the risk becomes concrete when people use generated text to decide whether a fee is valid, whether a promotion applies, or how to pursue a complaint.
The second situation is more consequential for businesses. A company that installs AI customer service is using the system as part of its own operating chain. If the tool explains a price, promise, return condition, or service procedure incorrectly, the association’s position is that the merchant cannot treat the mistake as though it came from an unrelated third party. It must supervise the system, make sure its information aligns with actual terms, and provide workable channels for human help and dispute resolution.
This is a useful shift in emphasis. The debate about AI policy often centers on foundation-model developers, training data, or the safety of highly capable systems. The association is focusing on a more ordinary but widespread setting: the customer trying to get a clear answer from an automated support window.
AI Customer Service Does Not Replace Merchant Duties
The new guidance does not say that companies should stop using AI in service operations. It recognizes that AI can make support faster and more convenient. The message is that automation cannot reduce the underlying duty to give consumers accurate, clear, and consistent information.
For businesses, that implies a need for governance beyond a polished chat interface. Product catalogs, price databases, promotion rules, contract terms, and after-sales policies must be kept current. When an AI system retrieves or generates an answer, companies need a way to detect when it has contradicted those sources. High-stakes interactions should not trap consumers in a loop with no route to a human employee who can correct the record.
The concern is especially relevant as companies put conversational systems into more complicated workflows. An AI tool can be useful for finding an order or explaining a basic service. It is far riskier when it speaks with confidence about financial claims, warranty exclusions, cancellation rights, or legal outcomes. The association calls for clearer risk reminders in areas where the consequences of a wrong answer may be greater.
China has already shown a willingness to tailor AI governance to particular use cases. EastFrontier’s coverage of the rules affecting AI companion features showed how regulators can focus on the risks of a specific form of AI interaction rather than treating every service as identical. The consumer-association guidance takes a comparable use-case approach but shifts attention from emotional reliance to commercial reliance.
It also complements the policy pressure on companies to add safeguards to consumer-facing model products. In a separate development, EastFrontier reported that the Media and Publishing Administration said ByteDance had added guardrails to Seedance. The issues are different, but both reflect a growing expectation that companies must manage what their AI systems do in real-world settings.
A Practical Test for China’s AI Commerce Boom
The real test will be implementation. Companies are likely to deploy AI customer service first where volumes are high and margins are thin: retail, travel, food delivery, telecoms, finance, and digital platforms. Those are also the places where an incorrect answer can affect a large number of consumers quickly.
The guidance suggests that a simple disclaimer will not be enough. A company cannot tell customers that the chatbot may be wrong, then use that warning to distance itself from an answer that shaped a purchase or service decision. The association’s reasoning is grounded in the nature of the relationship. If a tool is presented as the merchant’s customer service, it is part of the merchant’s service.
For consumers, the message is more cautious than alarmist. General AI can be useful for preliminary research and understanding a problem, but it should not replace authoritative records. Consumers should keep contracts, screenshots, official policy pages, and correspondence when making important decisions or pursuing a complaint. That is not a rejection of AI. It is an acknowledgement that a probabilistic language system is not the same thing as a binding promise.
The association’s notice therefore offers a concrete principle for the next stage of China’s AI rollout. Companies may automate conversations, but they cannot automate accountability away. As AI agents become more common in commerce, the quality of their answers will matter. So will the responsibility of the businesses that put them in front of customers.