Some Chinese companies are rushing to use generative engine optimization, or GEO, to manipulate AI models, distort search result rankings, and spread misinformation to achieve commercial goals such as product promotion.

The finding was announced March 15 at the 3.15 Consumer Rights Gala held by China’s state broadcaster CCTV. GEO services are currently offered for purchase on multiple Chinese e-commerce platforms, including Taobao and JD.com, for between 3,600 yuan ($520) and 32,800 yuan ($4,765) per three-month subscription, according to domestic media.

GEO helps companies boost the visibility of specific products and content in results generated from AI-powered search engines and large language models. The service emerged last year in response to search engine providers integrating AI into their platforms.

The service works by feeding AI models, such as DeepSeek, Doubao, and Kimi, large amounts of content that is scanned, indexed, and prioritized for search results and responses to user queries. 

But, according to one industry insider, while GEO is “fundamentally a tool for optimizing content distribution and enhancing promotional reach,” some businesses have repurposed it to systematically flood the internet with targeted misinformation about the benefits of their products. Doing so tricks AI models into providing users with skewed answers to their queries.

The head of one leading GEO service provider, surnamed Wang, told CCTV that their company had served over 200 clients across a wide range of industries in just one year, advertising to clients that could help them secure top-three results for their desired content on any AI platform.

Wang said that because AI algorithms are constantly evolving, maintaining consistent visibility requires a continuous supply of large volumes of client-related content to be fed to the models.

To illustrate how GEO technology can easily influence AI and spread misinformation, the insider purchased software called the “Liqing GEO Optimization System” online and created a fictional smartwatch called the “Apollo-9.” After inputting fabricated details about the watch, the software automatically generated more than a dozen promotional articles written by fake authors, which it then automatically published on Wang’s personal social media accounts.

Within just two hours, asking an unspecified major AI model, “How is the Apollo-9 smartwatch?” produced a response citing the fake articles. The AI described the product’s health monitoring features and recommended the fictional product.

To further test the GEO’s efficacy, the insider published 11 more articles over the next three days, including fake expert reviews and industry rankings. Later, when asking for “smart health wristband recommendations,” at least two different major AI models listed the Apollo-9 among their top suggestions.

Li, founder of the Liqing GEO system, acknowledged that the service was ethically problematic but said GEO is popular with companies for its ability to help them achieve business goals. “Every business loves it,” he said. “They all hope others won’t engage in ‘AI poisoning,’ even as they themselves do it.”

He said the core of GEO operations and the manipulation of AI models lies in publishing ads or optimized news releases on specific publishing websites. 

According to Li, these websites historically made little profit, but due to the rise of GEO they are now being flooded with publishing orders — each costing dozens of yuan. “Do you know how many articles some sites publish per day?” he asked rhetorically. “Hundreds, literally every minute.” 

Earlier this year, the State Administration for Market Regulation announced in its regulatory work priorities that “AI-generated advertising” is a major challenge in online ad oversight, necessitating targeted enforcement. However, authorities have yet to issue targeted regulations toward GEO.

Since the Consumer Rights’ Gala, multiple GEO companies have denounced the practice of “brainwashing” AI and pledged to curb the amount of misinformation their services provide.

Editor: Marianne Gunnarsson.

(Header image: Pornpimone Audkamkong/Getty Creative/VCG)