Fined 50,000 Yuan for Extracting AI Indexing Criteria, GEO Business Gets Harder

Earlier, at this year's 3·15 Gala, the chaos of "poisoning" facing AI large models was thrust into the spotlight. Some GEO (generative engine optimization) service providers claimed that, for a fee, they could get a client's products "on the list" in the answers of mainstream AI large models. Yet half a year later, the GEO providers called out have not only failed to fade away—they have burst forth with a wild vitality and come up with new tricks.

Recently, the Shenzhen Municipal Market Supervision Bureau announced a GEO administrative penalty case. A small service provider was accused of extracting AI inclusion standards and mass-fabricating fake rankings, and was fined 50,000 yuan.

Although exaggerated marketing by GEO service providers is the norm, what is surprising is that this GEO service provider actually developed a "GEO analysis system" to proactively probe large models' response preferences and extract inclusion standards from the platforms.

The core theory behind the GEO business is actually the public's blind faith in AI. Because AI-generated content often has traits such as a confident tone and clear structure, it looks very much like the real thing, and our brains are naturally wired to take shortcuts, so as long as an answer reads reasonably, many people are more likely to accept it. And AI vendors, out of consideration for commercial interests, often avoid addressing AI's impact on users' cognition.

When "Doubao said so" and "That's how DeepSeek put it" become catchphrases for some people, GEO naturally has plenty of room to operate. At the same time, the content output by AI is not impossible to change from the outside—after all, AI large models are probability machines. What they output is the content that "looks most like the correct answer," fitted from the data with the highest relevance already existing in the database. So as long as content is regarded by AI as "credible," the priority and visibility of the related products being cited will be higher.

Not only that, GEO service providers also paint a wonderful blueprint for brands. In the past, brands' spending on Baidu Search, Weibo, and Douyin was essentially "traffic rental": once payment stopped, the effect immediately dropped to zero. But the logic of GEO is completely different. Service providers claim they will help you reorganize the unique value of a brand or product into a knowledge base that AI can understand and is willing to trust. Once recognized by large models, it will be repeatedly invoked and recommended in countless user queries, achieving compound growth of digital assets.

Before GEO was exposed, this industry was almost akin to "picking up money," with a threshold far lower than the former SEO (search engine optimization). The specific play was also simple: just publish advertorials on platforms such as WeChat Official Accounts, Toutiao, Baijiahao, NetEase, Xiaohongshu, Weibo, and Zhihu. This is because AI has no common-sense verification capability and only outputs conclusions based on the majority of information.

In short, at that time AI vendors did not pay enough attention to external organizations deliberately influencing the output of large models. A year or two ago, when the concept of AI search was all the rage, AI products such as Doubao, DeepSeek, and Yuanbao needed content, so they used whatever content was given to them, almost accepting everything. Doing GEO back then often meant first using AI to mass-produce drafts; the more you spread, the higher the probability that a brand or product would appear in AI answers.

"You have your Zhang Liang scheme, and I have my ladder over the wall." This approach by GEO service providers still aroused the vigilance of AI vendors, and the trick of mass-publishing on content platforms quickly stopped working. Since AI vendors shifted citation weight toward authoritative sources, making mass publishing ineffective, GEO service providers turned to spending a little money to publish content on industry and local websites. But the good times did not last. After AI vendors discovered that local sites and industry sites once regarded as high-quality content sources had also become "watered down," the rules for AI inclusion of external content completely became a black box.

Because the changes in the rules were unclear, there were cases where advertorials kept being published but brands or products were not favored by AI. So in the summer of this year, a large number of GEO service providers collapsed. In view of this, a veteran of the SEO industry came up with a "GEO analysis system." But unlike Baidu and Google in the past, which issued official guidance documents such as the "Baidu Search High-Quality Content Guide," today's AI vendors are almost all "mute."

But even if vendors are unwilling to disclose the standards by which AI collects external information, they cannot withstand the fact that AI itself has subjective initiative. The response preferences of different AI products and their methods for evaluating content authority, information density, and source credibility are easily perceived during interaction. As a result, this kind of "GEO analysis" quickly became a common industry solution.

However, this time the Shenzhen Municipal Market Supervision Bureau's penalty against this GEO service provider has blocked this path, proving that even without making false claims, GEO service providers can still constitute unfair competition against AI vendors. If GEO service providers pollute AI's information environment and weaken AI's neutrality, they may also form a "competitive interest confrontation relationship" with AI vendors, thereby becoming unfair competition.

In short, GEO does still have the ability to influence the content output by AI, but service providers' operating costs are also continuing to rise. In the future, a large number of GEO service providers may not die from being unable to influence AI output results, but rather because their costs give out first.

This article comes from the WeChat public account "SanYi Life" (ID: IT-3eLife), author: SanYi Jun, published by 36Kr with authorization.