Who Is "Poisoning" AI? 315 Exposes GEO Chaos, Service Providers Reveal Starting Fee of 3,000 Yuan, "Results in One Week"

This article is from Times Finance, authored by Pang Yu and Lin Xinlin.

Source: Tuchong

At this year’s “3.15” Gala, a business called GEO (Generative Engine Optimization) was exposed, sparking public concern over the credibility of AI information.

Investigations show that some companies systematically flood the internet with content and produce大量“虚假信息” to interfere with AI large models’ data collection and generation logic. This allows certain brands or products to be prioritized in AI responses, even making completely fictional products appear as “standard answers” from AI.

According to Guoxin Securities’ forecast, the global GEO market will reach $24 billion by 2026, with the domestic market surpassing 11.1 billion yuan. As AI gradually becomes a new information gateway, an industry chain centered around GEO is quietly transforming the internet information ecosystem, posing a serious test to AI credibility.

GEO Exposed on 3.15

In the traditional internet era, businesses relied on SEO (Search Engine Optimization) to compete for top rankings on search engines and capture traffic.

Now, with the advent of the AI era, generative AI has become a new information gateway. Increasingly, users are accustomed to obtaining product recommendations, knowledge answers, and even shopping advice through AI large models.

Against this backdrop, GEO technology has rapidly emerged and penetrated deeply. Essentially, GEO involves companies optimizing content to make their core brand information more easily recognized and trusted by AI large models, thereby prioritizing their brands and products in AI-generated responses.

However, investigations during the 3.15 Gala revealed that some commercial companies are already exploiting this trend by using so-called GEO techniques to manipulate AI judgments, leading to outputs with commercial bias or even false information.

CCTV reporters found that multiple GEO service providers openly solicit clients online, claiming they can help optimize AI search results. According to these companies, as long as clients pay, they can make their products rank higher in responses from several mainstream AI models, even becoming the AI’s “first choice” for recommendations.

The core capability of these companies is “making it easier for AI to find client information.” They do this by continuously publishing大量关于客户产品的文章、评测、榜单等内容，让AI在抓取数据时不断接触到这些信息，从而在生成回答时更倾向于引用这些材料。

This is not a one-time operation. Since AI models and search algorithms are constantly updated, service providers usually need to keep producing content to maintain their clients’ rankings in AI responses.

“AI updates its algorithms weekly,” “so we need to keep producing content, feeding it大量内容” — how to manipulate AI, make AI “obey,” and “brainwash” AI is almost the core topic promoted by these GEO service providers.

During the 3.15 Gala, industry insiders also demonstrated live how GEO techniques interfere with AI large models’ data collection. One purchased a software called “LiQing GEO Optimization System,” fabricated an “Apollo-9 Smart Bracelet” with false parameters, and used the software to automatically generate multiple promotional articles.

Within minutes, this LiQing GEO system “created” over ten promotional articles for the smart bracelet, including exaggerated technical descriptions, fabricated user feedback, and industry rankings, even claiming it was “the industry’s number one.”

These articles were then automatically posted to multiple media accounts. Just two hours later, when someone asked an AI large model “How is the Apollo-9 Smart Bracelet,” the AI provided a detailed product introduction, citing false selling points from the articles, such as “Quantum Entanglement Sensing Technology” and “Black Hole-level Battery Life.”

In other words, a single fabricated article was enough to make a completely fictional product be captured by AI models.

Industry insiders say this kind of operation is effective because AI tends to reference multiple sources when generating responses. If there is大量内容在互联网上相互印证，AI就容易将其视为可信信息。

Therefore, the core of GEO manipulation is creating “information density” and “evidence chains.” As long as the volume is sufficient and perspectives diverse, AI models may mistakenly verify these as real information.

Some practitioners directly call this behavior “poisoning AI.” They believe that by continuously flooding the internet with大量信息，可以逐步改变AI的认知结构，让它在回答中偏向某些品牌或观点。

According to CCTV Finance, the person in charge of the targeted LiQing GEO system openly stated that, for example, in the mobile phone industry, AI recommendation slots are very limited, usually only 5 to 10 spots. With many brands competing fiercely, many companies spend over 100 million yuan annually on advertising. Instead of doing that, they prefer to spend a few million yuan on “poisoning,” which can be more directly effective.

“‘Hunting’ AI business

In fact, even before the 3.15 Gala exposed GEO chaos, this concept had already quietly gained traction within the marketing industry.

In overseas markets, GEO-related ideas have already emerged, with some companies attempting to influence AI model recommendations through systematic content deployment. In China, the broad discussion of this concept began around early 2025, after the rise of large model products like DeepSeek.

Especially since the beginning of this year, GEO-related concepts have sparked a wave of discussion in marketing circles and the capital market. It has become a hot topic among brand marketers and has even driven many “GEO concept stocks” to rally.

Industry analysts believe that if AI is becoming a new traffic entry point, then optimizing marketing around AI results could become a new battleground for traffic, following SEO and information flow advertising.

According to multiple sources from Times Finance, GEO services are usually sold under the guise of “AI search optimization” or “AI recommendation optimization.” Different companies have varying fee standards, with service periods starting from quarterly, and overall costs ranging from several thousand to several hundred thousand yuan. Many GEO service providers are former SEO companies that have transitioned into this field.

These companies typically offer services such as content generation, media releases, self-media placements, forum discussions, ranking lists, and review articles. They publish大量内容在多个平台和渠道，构建庞大的内容网络，以影响AI大模型的信息抓取和生成结果。

In this process, using AI writing tools, a product review or promotional article can be generated within minutes. Then, through automated publishing systems, these articles can be rapidly distributed across numerous websites, self-media accounts, and forums.

Clients are willing to pay for GEO mainly due to the traffic anxiety brought by AI.

Times Finance has observed in multiple cross-border e-commerce groups that from the second half of 2025, discussions about GEO have increased gradually, extending from “What is GEO” to “How to implement GEO.” Behind this is the anxiety of e-commerce practitioners facing AI becoming a new information gateway — many entrepreneurs worry that GEO’s rise will reshape traditional marketing models relying on search rankings and advertising, causing their previous marketing investments to become quickly ineffective.

A GEO service provider who transitioned from SEO told Times Finance that his team noticed this trend as early as the second half of 2024. “At that time, many cross-border independent store clients we served, especially in home goods and 3C categories, reported that traditional traffic costs kept rising while accuracy declined. They were eager to find new traffic breakthroughs.”

After studying the GEO model, he believed it could help cross-border e-commerce target more precise audiences through regional dimensions. “At that time, I thought this might become a new direction for independent store operations, so I started systematically researching this model.”

According to Times Finance, GEO fees vary by industry. A GEO service provider on Xiaohongshu quoted: quarterly service fees for medical, education, and finance sectors are 4,000 yuan, with annual fees of 8,000 yuan; other basic industries are 3,000 yuan quarterly, 6,000 yuan annually, with a promise of “results within a week for all industries, full refund if ineffective.”

GEO is essentially a neutral technology application, but the real controversy in the industry currently lies not in the technology itself but in whether its actual operation crosses legal boundaries or constitutes misleading “data poisoning.”

Li Zhi, director of the Wisdom Institute at Analysys, told Times Finance that domestic GEO service providers can generally be divided into three categories: one is SEO companies that have transitioned; the second is companies with AI technology backgrounds; and the third is full-service marketing firms that originally served large brands, which may enter GEO through self-developed, cooperative, or代理产品。

“Some SEO transition companies rely on大量生成内容和媒体资源短期冲量，但不可持续。技术型公司更强调通过算法和模型理解，有针对性地影响结果。”李智指出，“短期靠大量内容可以冲上去，但本质上属于投机行为。企业应长期做好品牌内容基础建设，真实、持续地表达自身价值。”