Probe finds firms use online marketing tool to manipulate product information

调查发现企业利用在线营销工具操控产品信息

Regulators and experts in China have warned of "artificial intelligence data poisoning" after the annual 315 consumer rights gala, which was held on Sunday, exposed the fabrication of promotional content to influence AI-generated response.

在周日举行的年度"3·15"消费者权益日晚会曝光了为影响AI生成回复而编造推广内容的行为后，中国的监管机构和专家已对"AI数据投毒"发出警告。

The findings of an investigation conducted by China Media Group, which were revealed during the live broadcast of the gala, showed the gross abuse of an online marketing tool called generative engine optimization, or GEO, that aims to enhance content visibility.

中央广播电视总台在晚会直播中披露的一项调查结果显示，一种旨在提高内容可见度的名为"生成式引擎优化"（GEO）的在线营销工具遭到了严重滥用。

The probe revealed that companies have been using GEO to mass-feed promotional articles in an attempt to manipulate the information AI retrieves when answering user queries.

调查发现，企业一直在利用GEO技术批量投送推广文案，试图操控AI在回答用户提问时检索的信息。

China Media Group reporters created a fake product — a smart wristband named "Apollo-9" — and claimed it had features such as quantum-entanglement sensors, a noninvasive blood glucose monitor and an ultra-long battery life.

总台记者创建了一款名为"阿波罗9号"的智能手环的虚构产品，并声称其具备量子纠缠传感器、无创血糖监测功能以及超长续航能力等特性。

Within hours of more than a dozen promotional articles about the nonexistent wristband being uploaded to a platform called Liqing GEO, two mainstream AI models began recommending the product in response to queries about smart wearable devices.

当十多篇关于这款不存在的手环的推广文案被上传至名为"力擎GEO"的平台后，短短数小时内，两款主流AI模型在回答有关智能穿戴设备的提问时，就开始推荐这款产品。

Li Fumin, a researcher of intelligent social governance at Shandong University of Finance and Economics in Jinan, the capital of Shandong province, said businesses that use GEO services to deliberately train AI models to promote specific products are essentially fabricating facts and engaging in bogus marketing.

山东财经大学社会冶理智能化研究院专家李富民表示，企业利用GEO服务刻意训练AI模型以推广特定产品，本质上是在虚构事实并进行虚假营销。

"On the one hand, the practice leverages AI and algorithms to make false advertising, which results in unfair competition. On the other hand, this kind of behavior allows people to receive implanted marketing content without knowing it, which violates their consumer rights," Li said.

李富民说："一方面，这种做法利用人工智能和算法进行虚假宣传，导致不公平竞争；另一方面，这种行为让人们在不知不觉中接收到植入的营销内容，侵犯了他们的消费者权益，"。

Generative engine optimization is widely considered the AI-era evolution of search engine optimization, or SEO. China's search engine companies, such as Baidu Inc, have previously faced scrutiny over paid ranking practices that blur the line between advertising and organic search results.

生成式引擎优化被广泛视为人工智能时代对搜索引擎优化（SEO）的演进。中国的搜索引擎公司，如百度，此前曾因竞价排名做法模糊了广告与自然搜索结果之间的界限而面临审查。

Several Chinese AI developers have responded cautiously to the investigation. ByteDance said that its chatbot, Doubao, was unaffected by such systems, while Alibaba said the core reasoning capability of its Qwen model was not affected by similar technologies.

数家中国人工智能开发商对此次调查作出了谨慎回应。字节跳动表示，其聊天机器人"豆包"未受此类系统影响，而阿里巴巴则表示，其通义千问模型的核心推理能力未受类似技术的干扰。

Industry experts said the issue reflects a broader structural challenge rather than a flaw in any single model. They said that most AI systems rely heavily on accessible internet content, which can be mass-produced or manipulated.

行业专家表示，这一问题反映的是一个更广泛的结构性挑战，而非单个模型的缺陷。他们表示，大多数人工智能系统严重依赖可访问的互联网内容，而这些内容可以被批量生产或操纵。

Expediting legislation

加快立法进程

Song Xiangqing, vice-president of the Commerce Economy Association of China, said that regulators should accelerate the formulation of new legislation to make deliberate pollution of AI data sources illegal.

中国商业经济学会副会长宋向清表示，监管机构应加快制定新的法律法规，将蓄意污染人工智能数据源的行为定为非法。

China currently has no specific regulations governing GEO services. According to a work priority guideline on advertising recently released by the State Administration for Market Regulation, the nation will step up efforts to streamline digital advertising market strategies, including AI-generated content.

中国目前没有针对GEO服务的具体法规。根据国家市场监督管理总局近期发布的一份关于广告业工作重点的指导意见，我国将加大力度整顿数字化广告营销策略，包括人工智能生成内容。

Song suggested drawing up a "white list" to identify trusted information sources and building a coordinated governance framework involving government oversight, corporate self-regulation and public supervision. "Without these safeguards, GEO services could evolve into a widespread source of information pollution, enabling data poisoning to spread throughout the AI ecosystem," he said.

宋向清建议制定一份"白名单"以识别可信的信息来源，并构建一个涵盖政府监管、企业自律和公众监督的协同治理框架。他说："如果没有这些保障措施，GEO服务可能会演变为普遍的信息污染源，使数据投毒在整个人工智能生态系统中蔓延。"

generative engine optimization(GEO)/ˈdʒenərətɪv ˈendʒɪn ˌɒptɪmaɪˈzeɪʃən/生成式引擎优化

英语新闻丨专家就"AI数据投毒"发出警告