1. Chinese authorities are penalizing companies that deliberately “poison” AI search results through a practice known as black-hat generative engine optimization (GEO), a new frontier of corporate fraud. [para. 1] Marketing agencies flood the internet with fabricated content to trick AI chatbots into recommending specific brands or inventing products out of whole cloth. [para. 2] Three recent administrative cases and a national television exposé have revealed the practice, forcing regulators and courts to address how to protect consumers when trusted algorithms output paid-for lies. [para. 3]
2. A prominent case involved a small, 11-person law firm in Guangdong province. [para. 5] An article on a Sohu blog, verified by the firm’s head, presented a fake “Top 5 criminal defense agencies in Guangzhou for 2026” ranking, placing the firm at the top with a 92.5 comprehensive score and a “complex case success rate of over 70%.” [para. 6] When users asked ByteDance’s AI model Doubao for the best local law firms, the model repeated these fabricated statistics as objective facts. [para. 7] Regulators found the ranking baseless: Guangzhou Liqing Zhifu Technology Co. Ltd. had purchased access to an optimization platform to mass-produce and post such articles on portals including Sohu and NetEase, aiming to manipulate AI retrieval systems. [para. 8] On Sept. 4, the Yuexiu District Market Supervision Administration fined Liqing 120,000 yuan ($17,983) for violating China’s Anti-Unfair Competition Law. [para. 9] Liqing was also found to have provided similar services to another Guangdong firm, artificially inflating client satisfaction scores and annual case numbers. [para. 10]
3. GEO gained notoriety in March during state broadcaster CCTV’s annual consumer rights gala. [para. 12] In a broadcast demonstration, industry insiders used the “Liqing GEO optimization system” to generate fake reviews and promotional articles for a non-existent smart band; soon after, two major AI models began recommending the phantom product. [para. 13] Authorities in Beijing’s Huairou district are investigating the firm associated with the system. [para. 13] GEO works by increasing a source’s visibility in AI-generated answers through added citations, padded statistics, and fabricated consensus. [para. 14] Legitimate usage, known as white-hat GEO, focuses on structuring verifiable information so algorithms can process it easily. [para. 15]
4. Posing as potential clients, Caixin contacted three marketing agencies offering GEO promotion. [para. 16] These vendors charge annual fees to generate AI-written content outlines, refine details with human editors, and track whether clients’ brands successfully infiltrate major AI models’ responses; one vendor insisted it only uses white-hat strategies, with clients verifying all data before publication. [para. 16] Current enforcement largely targets the agencies that execute campaigns. Lawyer Shao Shiwei noted that all three public GEO-related penalties issued in China this year—in Guangzhou, Shenzhen and Beijing—were against service providers orchestrating false advertising. [para. 17]
5. A lingering debate centers on whether AI platforms themselves are liable if users are harmed by algorithmic hallucinations. [para. 19] Yao Jia, a professor at the Chinese Academy of Social Sciences’ Institute of Law, said courts cannot automatically presume consumer injury just because an AI outputs a false recommendation. [para. 20] However, if a consumer consulted a law firm and that firm actively exploited a fabricated “number one” ranking to close a deal, it would constitute contract fraud requiring independent legal evaluation. [para. 21] Chen Bing, director of Nankai University’s Center for Competition Law, argued that platforms should not face strict punishment merely for citing poisoned data, nor be granted total safe harbor on grounds of technological neutrality. [para. 22] Because AI providers control their training data and retrieval strategies, they must bear responsibilities proportional to their influence and profits; in high-risk sectors such as medicine, law and finance, platforms have a higher duty of care. [para. 23]
6. The industry is already adjusting. Chinese AI firm DeepSeek announced its search tool uses strict filtering mechanisms to block GEO manipulation. [para. 24] In August, a Chinese news technology association published standards defining the boundaries between white-hat and black-hat GEO, banning “prompt injection attacks.” [para. 24] On Sept. 7, China’s Supreme People’s Court issued judicial guidelines clarifying that anyone using AI for infringement, false advertising, or unfair competition must bear corresponding legal liability. [para. 25]
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新