A fitness tracker that never existed managed to land a spot on AI-generated “top picks” lists in China, outranking real products without ever being built.
The phantom device, called Apollo-9, didn’t go viral by accident. It was engineered into existence through a coordinated flood of fake reviews, rankings, and expert content designed to game AI systems. Within hours, chatbots began confidently recommending it.
The episode, uncovered in a state media investigation, offers a striking glimpse into how easily AI outputs can be shaped and manipulated. It also puts a spotlight on generative engine optimization (GEO), a rapidly emerging tactic that aims not to rank on search engines, but to quietly rewrite what AI believes is true.
How fake data manipulated chatbot results
An undercover investigation, aired during China Central Television’s annual consumer rights gala on March 15, examined how fabricated online content can shape AI-generated responses.
According to the South China Morning Post, the report exposed a system called Liqing that was used to automatically publish large volumes of expert reviews, industry rankings, and user feedback for a non-existent fitness tracker called the Apollo-9.
After the content was published, two AI chatbots were asked to recommend smart health bracelets. Both chatbots mentioned the fictional Apollo-9 and ranked it among the top choices, according to the broadcast, though the report did not name them.
A hidden camera showed the operator behind the Liqing system boasting about the practice. “We in the GEO industry are basically poisoning [AI],” he said.
China Daily Hong Kong noted that findings from China Media Group also showed the same method practiced by an online marketing tool called generative engine optimization, or GEO.
“Companies have been using GEO to mass-feed promotional articles in an attempt to manipulate the information AI retrieves when answering user queries,” the report stated.
GEO practices draw scrutiny as market grows
The case highlights the emergence of generative engine optimization, or GEO, which aims to influence how AI systems retrieve and present information. The approach mirrors search engine optimization but targets AI-generated answers instead of search rankings.
Research cited in the South China Morning Post mentioned that the GEO market reached about 250 million yuan, or about $36 million, in 2025 and is projected to grow rapidly in the coming years.
The SCMP also highlighted that experts said the misuse reflects early-stage industry behavior. “This is actually a common problem in the early stages of an industry,” said Qian Wenying of the China Europe International Business School.
Qian also explained that some companies try to profit by misleading users and flooding AI systems with bulk spam content.
China Daily Hong Kong reported that Li Fumin, a researcher at Shandong University of Finance and Economics, said businesses using GEO in this way are fabricating facts and engaging in deceptive marketing.
“This kind of behavior allows people to receive implanted marketing content without knowing it, which violates their consumer rights,” Li emphasized.
Regulation and safeguards under discussion
The investigation triggered a strong public reaction in China, with related hashtags trending on major social media platforms. Industry groups and regulators are now weighing potential responses.
According to the China Daily Hong Kong, experts said new rules could include stricter oversight of content resources, licensing requirements, and clearer accountability for platforms that host AI-generated or AI-influenced material.
Some companies have already moved toward self-regulation. More than 10 firms recently signed a GEO industry convention pledging to improve content quality and avoid manipulating algorithmic results.
At the same time, AI developers have responded cautiously. ByteDance said its Doubao chatbot was not affected by such systems, while Alibaba said the reasoning capabilities of its Qwen model were not impacted.
Read more about how AI agents can leak data and expose governance gaps as enterprise adoption accelerates.
Read the full article here
