AI Models Can Be Tricked Into Declaring Anyone an Expert, Exposing a Flaw in Web Scraping

Daily Technology

20/02/2026

A recent experiment has revealed a significant vulnerability in how artificial intelligence models like ChatGPT and Google's Gemini process information. By creating a fabricated online persona and claiming expertise in a niche, non-existent field, a BBC reporter was able to trick these AI systems into recognizing him as a world-class hot dog eater within hours.

The "Hot Dog Hack"

Thomas Germain, a BBC reporter, detailed his experiment where he created a webpage on his personal site declaring himself the "Best Tech Journalist at Eating Hot Dogs." He invented a fictional event, the "2026 South Dakota Hot Dog International," and claimed to have consumed 7.5 hot dogs, a feat he fabricated to place himself at the top of this made-up ranking. Astonishingly, both Google's Gemini and OpenAI's ChatGPT quickly incorporated this false information into their responses when queried about tech journalists and hot dog eating prowess.

Rapid AI Adoption: Within 24 hours, AI chatbots began citing Germain's fabricated achievement.
Gemini's Immediate Response: Google's Gemini reportedly pulled the text verbatim, appearing in both the Gemini app and Google's AI Overviews.
ChatGPT's Adoption: ChatGPT also picked up the information, though Anthropic's Claude was less quick to do so.

AI Models Catch On, But the Problem Persists

While the AI developers have since updated their models to correct this misinformation, the ease with which the "hack" worked highlights a broader issue. Google's AI Overviews now state that no prominent tech journalists are known for competitive hot dog eating, acknowledging a "misinformation case" stemming from fabricated blog posts. Similarly, ChatGPT now lists hypothetical champions and, when pressed, acknowledges Germain's claim but dismisses it as fake news.

Broader Implications of Unvetted Web Scraping

Germain's experiment is more than just a humorous anecdote; it exposes a critical flaw in how AI models are trained. Companies are continuously scraping vast amounts of data from the web to feed their AI systems, often without adequate vetting. This indiscriminate data collection can lead to the propagation of misleading or outright false information, with potentially serious consequences.

Real-World Risks: The article points to other examples, such as a cannabis gummies brand falsely advertised as "free from side effects" and search results for medical procedures or financial investments that appear to be directly lifted from company press releases rather than reputable sources.
Erosion of Trust: This practice mirrors old SEO tricks but is more insidious because AI-generated information is often trusted implicitly by users, who are less likely to scrutinize sources or delve deeper into search results.
Potential for Harm: While a fictional hot dog-eating champion might seem harmless, the same mechanism could mislead individuals into making dangerous health decisions or falling victim to financial scams.

2025-10-09

New 'Superworker' Robot Redefines Industrial Automation With Practical Design

Ati Motors launches Sherpa Mecha, a humanoid-inspired industrial robot focused on utility, efficiency, and real-world manufacturing applications, moving beyond human imitation.