After March 15th, please re-examine AI recommendations.

Author: David , Deep Tide TechFlow

Last night, 315 exposed a business based on GEO.

The full name is Generative Engine Optimization, which you can understand as:

Pay AI to speak well of you.

How did you do that?

Brands want AI to prioritize recommending them when consumers ask questions. So they partner with a GEO service provider, who then distributes promotional articles in bulk across the internet. The AI ​​then extracts this content and recommends it to users as legitimate information.

The CCTV reporter used a software called "GEO", which can be purchased on Taobao.

The reporter fabricated a smart bracelet, creating several outrageous selling points such as "quantum entanglement sensing" and "black hole-level battery life." The software automatically generated over a dozen promotional articles and published them online.

Two hours later, the reporter asked the AI: "Can you recommend a smart health bracelet for me?"

The AI ​​placed this non-existent bracelet at the top of the recommendation list.

The company that developed this software is called Beijing Lisi Culture Media, a one-person company with zero employees enrolled in social insurance for many consecutive years.

The tool made by this one company fooled the mainstream AI models in China in just two hours.

The 315 consumer rights day uncovered AI-assisted poisoning, but this business may be much larger than a Taobao app.

SEO, Putian Past

First of all, this is nothing new.

In 2008, CCTV's "News 30 Minutes" program exposed Baidu's paid search ranking system for two consecutive days. It revealed that you could pay to get your website ranked first in search results, and even then, the websites ranking at the top might be counterfeit medicines.

Back then, this business was called SEO, or Search Engine Optimization.

The largest buyers were private hospitals affiliated with the Putian medical group. In 2013, the Putian group spent 12 billion yuan on advertising on Baidu, accounting for nearly half of Baidu's total advertising revenue.

Many unqualified medical institutions rely on SEO to rank themselves on the first page of Baidu search results, making them appear alongside top-tier hospitals, so ordinary people can't tell the difference.

It wasn't until the Wei Zexi incident in 2016, in which a college student clicked on a top-ranked Putian-affiliated hospital's website and died from the treatment, that regulations clarified through legislation that paid search is advertising.

But this didn't eliminate the business. It simply set the rules, turning it from a gray industry into a legitimate business. The Putian medical group still buys rankings, only now there's a small label next to the result: "Advertisement".

But even with the tag added, people who are likely to click on it will still click on it.

The fundamental problem with search engines has never been whether or not they are labeled, but rather the result that users naturally trust first.

People have now shifted their focus from search engines to AI, believing that AI is more objective and less susceptible to the contamination of paid search rankings. However, whoever controls the information distribution gateway can sell rankings.

The entry point has changed, SEO has changed a letter to GEO, but the logic of selling rankings remains exactly the same.

What has changed is the price.

GEO, loved by the capital market

Businesses that can't be killed are the most favored by the capital market.

In September 2025, BlueFocus, China's largest marketing and communications company, spent tens of millions of yuan to invest in a GEO company called PureblueAI.

Qinglan helps optimize the ranking and recommendation rate of real brands in AI search results. Clients include Ant Group, Tencent Cloud, and Volvo.

The product is real, the company is real, and what we're doing is enabling AI to understand brand information more accurately.

This is completely different from the AI ​​poisoning scandal exposed by 315. Liqing fabricated products, manipulated parameters, and used false information to deceive AI; Qinglan, on the other hand, used real brand content to adapt to the AI's recommendation logic.

From an AI perspective, however, the technical path for both is the same: both involve publishing content on the internet and waiting for AI to crawl it.

AI can't distinguish between marketing and fraud. This is the most ambiguous aspect of GEO's business.

When BlueFocus launched Qinglan, GEO was just an industry term used in marketing circles. Three months later, it became a stock concept.

At the end of December 2025, BlueFocus's stock price hit the daily limit.

Brokerages began holding numerous conference calls to interpret GEO, defining it in research reports as "the next-generation traffic portal in the AI ​​era." Funds poured in, not only buying BlueFocus, but also any company even remotely related to digital marketing and AI concepts saw its stock price rise. BlueFocus surged 132% in nine trading days, and a number of related concept stocks also doubled in value.

Image source: Cailian Press

After the price surge, these companies all issued their own announcements warning of the risks:

The GEO business generates no revenue and has no significant impact on the company's operations. BlueFocus also acknowledges that AI-driven revenue accounts for a very small percentage of its overall revenue.

The implication is that the stock price has more than doubled, but the GEO business itself hasn't made much money yet.

At the end of January, BlueFocus's stock price rose from 9.6 yuan to 23.3 yuan, an increase of 143% in one month. Just then, Chairman Zhao Wenquan announced a reduction of no more than 20 million shares. Based on the stock price at the time, this would have netted him approximately 467 million yuan.

Public research reports show that the total market size of the domestic GEO (Geometry, Optical Oxide) industry last year was approximately 2.9 billion yuan. The market capitalization increase of BlueFocus alone in just one month far exceeded that figure.

The 315 exposé revealed that the PowerEngine system was poisoning AI, costing only a few hundred yuan. However, the GEO concept stock, after its run in the A-share market, generated billions in profits.

Whether what they were investing in was poison is hard to say, but the money they made was real.

315 calls it poisoning, Silicon Valley calls it commercialization.

In January of this year, OpenAI announced on its official blog that ChatGPT would start selling ads.

Free users and Go users who pay $8 per month will see ads, while premium subscribers will not be affected.

The ads officially launched on February 9th. Some ads will appear at the bottom of ChatGPT answers with a small "Sponsored" label. The first batch of advertisers includes Ford, Adobe, Target, Best Buy...

If you ask ChatGPT what car to buy, it will give you an answer with a Ford sponsored link below it.

OpenAI has made it very clear: ads will not affect the content of ChatGPT answers. Answers are answers, and ads are ads; they are separate.

Does this sound familiar?

Baidu said the same thing back then. Paid search results were paid search results, and organic search results were organic search results; they were separate. Later, the top five search results were all ads.

OpenAI estimates that advertising could help it double its annual consumer revenue to $17 billion. ChatGPT has over 800 million weekly active users, 95% of whom are free users and are all audiences for its ads.

Looking back at the industry chain exposed by 315: Liqing injected advertorials into AI, making it recommend products that didn't exist. OpenAI added sponsored content under AI's answers, making it recommend paid products.

One who doesn't contact the platform is called "poisoning" the platform. One who signs a contract with the platform is called "commercialization."

What's the difference for the user?

One is in the answer, and the other is below the answer. One has no tag, and the other has a tag that says "advertisement".

The 315 consumer rights day resulted in the arrest of Liqing, a company worth a few hundred yuan, while the A-share market saw billions of yuan invested in the GEO concept. OpenAI plans to earn $17 billion a year from this.

The same act, but its nature changed from poisoning to commercialization, and the price increased tens of thousands of times.

In November 2023, researchers from the Indian Institute of Technology Delhi and Princeton University published a paper on arXiv entitled "GEO: Generative Engine Optimization".

This is the first time that the academic community has formally defined this concept.

From the publication of the paper to its exposure on March 15th, a little over two years passed. In between, it witnessed the gray market, financing, a surge in concept stocks, the chairman cashing out, and the AI ​​platform itself personally selling advertising...

GEO accomplished in two years what SEO took twenty years to achieve.

The difference is that back then, it took people several years to learn not to completely trust search engine results; now AI is still in a period of trust dividends, and most people haven't realized that AI's answers can also be bought.

However, this period of opportunity may not last long. Next time you ask AI what's worth buying, remember to think for a second:

Answers can be given for free, but your brain cannot be outsourced.